Whoever Holds the Keys Holds the Coins

Crypto hacks took $3.4 billion in 2025, with the $1.5 billion Bybit breach alone accounting for 44%, while attacks on individual wallets rose. The difference between hot and cold wallets, and the habits that protect your coins.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

This article explains crypto security basics. It is general information, not financial or investment advice.

In traditional banking, a bank error or a stolen card can often be reversed. In crypto, whoever controls the private key controls the funds, and a confirmed transfer generally cannot be undone. Security is not a feature you buy; it is a set of habits.

The Scale of Theft

According to Chainalysis, crypto hacks totalled $3.4 billion in 2025.

• The Bybit exchange breach on 21 February 2025 took nearly $1.5 billion in ether — around 44 percent of the year's total and the largest digital heist in crypto history.
• Hackers linked to North Korea stole $2.02 billion in 2025, including the Bybit theft, attributed to a group known as TraderTraitor.
• North Korean attacks made up a record 76 percent of service compromises.
• The top three hacks represented 69 percent of losses from services.

Chainalysis also reported that attacks on individual wallets were rising. Exchanges are not the only target.

What a Wallet Actually Holds

A crypto wallet does not hold coins. The coins exist on the blockchain. A wallet holds the private keys that prove you can move them.

Most wallets generate a seed phrase — usually 12 or 24 words — from which all the keys can be recreated. Anyone with that phrase can take everything. Anyone who loses it, with no backup, may lose access forever.

Hot Wallets

A hot wallet is connected to the internet: a phone app, a browser extension, or an account on an exchange.

Strengths: convenient for frequent transactions and trading.

Weaknesses: exposed to malware, phishing sites, malicious browser extensions and compromised devices. On an exchange, the exchange holds the keys — which means its security, not yours, protects the funds.

Cold Wallets

A cold wallet keeps private keys offline. The most common form is a hardware wallet, a small device that signs transactions without exposing the keys to the connected computer.

Strengths: far harder to attack remotely.

Weaknesses: less convenient, and the physical device and seed phrase must be protected from loss, damage and theft.

A Sensible Split

Many holders use both: a small hot-wallet balance for spending or trading, and the bulk in cold storage — the way people keep some cash in a wallet and savings elsewhere.

Habits That Prevent Most Losses

1. Never type your seed phrase into a website, app or message. No legitimate support agent will ask for it.
2. Write the seed phrase down offline and store it securely. Do not photograph it or save it in cloud notes.
3. Buy hardware wallets only from the manufacturer or an authorised seller — never second-hand.
4. Check the address before sending. Malware can swap a copied address for the attacker's.
5. Use an authenticator app, not SMS, for exchange accounts.
6. Be suspicious of "wallet verification" or "airdrop claim" links. They are a common way wallets are drained.
7. Test with a small amount first when sending to a new address.

The Exchange Question

Leaving funds on an exchange means trusting its security and solvency. Regulation is increasingly setting standards for how providers operate — in the European Union, MiCA now requires licences — but regulation does not make a hack impossible, as Bybit showed.

Holding coins yourself removes that dependency and adds responsibility. A regulated fund such as a spot Bitcoin ETF removes both the keys and the responsibility, at the cost of not owning coins directly. There is no option without trade-offs.

Related reading

• FBI: $11.4 Billion Lost to Crypto Scams in 2025, Over Half of All Fraud
• What Moves Exchange Rates, and Why the Fed's 16 September Decision Matters
• Pig Butchering Scams: The Long Con Now Run by AI
• Bitcoin ETFs Explained: $5.4bn Out in Six Months, $3.5bn Back in August

Sources

• "Crypto hacks hit $3.4 billion in 2025, attacks on individual wallets rise: Chainalysis," The Block — theblock.co
• "North Korea-linked hackers steal $2.02 billion in 2025, leading global crypto theft," The Hacker News — thehackernews.com
• "Collaboration in the wake of record-breaking Bybit theft," Chainalysis — chainalysis.com
• "Crypto theft in 2025 concentrated in fewer, larger breaches," BankInfoSecurity — bankinfosecurity.com

Bitcoin Now Trades Like a Stock Fund. The Money Moves Like One Too.

US spot Bitcoin ETFs had their first negative half-year in 2026 with $5.4 billion of outflows, then drew $3.5 billion in August and $730.9 million on 3 September. How these funds work and what the flows show.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how Bitcoin ETFs work. It is general information, not financial or investment advice.

For most of Bitcoin's history, owning it meant opening a crypto exchange account and managing a wallet. Spot Bitcoin exchange-traded funds changed that for mainstream investors.

What a Spot Bitcoin ETF Is

An exchange-traded fund is a fund that trades on a stock exchange like a share. A spot Bitcoin ETF holds actual bitcoin, so its price tracks bitcoin's market price.

Buying a share of the ETF gives exposure to bitcoin's price through an ordinary brokerage or retirement account. The fund, not the investor, holds the coins with a custodian.

What You Gain and Give Up

Gained: no wallet or private keys to manage, familiar brokerage accounts, and the regulatory framework that applies to listed funds.

Given up: you do not own bitcoin directly, cannot send it to anyone, pay an annual management fee, and can only trade when the stock market is open — while bitcoin itself trades around the clock. Direct ownership brings its own security responsibilities, covered in our wallet security guide.

The Market Leader

BlackRock's iShares Bitcoin Trust (IBIT), launched in January 2024, is the largest spot Bitcoin ETF by assets.

IBIT held about $54 billion in assets in March 2026 — close to 49 percent of the US spot Bitcoin ETF market — and around $67 billion by early May. Asset figures move with bitcoin's price as well as with investor flows, which is why reported totals vary by date.

In Q1 2026, IBIT captured 47 percent of spot Bitcoin ETF inflows, ahead of Fidelity's FBTC at 32 percent and Grayscale's GBTC at 11 percent.

2026: Money Out, Then Money In

The year has not moved in one direction.

• First half of 2026: $5.4 billion in net outflows — the first negative half-year since the products launched.
• August 2026: $3.5 billion in net inflows.
• 3 September 2026: $730.9 million of inflows in a single day — the largest since 14 January — with IBIT taking about $454 million, roughly 62 percent.

What the Flows Tell You

ETF flows show when large and mainstream investors add or reduce exposure. Sustained inflows mean fresh money buying bitcoin through funds; outflows mean redemptions.

What flows do not do is predict prices. Money often follows price moves rather than leading them, and a single strong day says little about the next month.

Flows also respond to wider forces, including interest rate expectations — relevant ahead of the Federal Reserve's 16 September decision.

The Risks Do Not Change

An ETF wrapper changes how bitcoin is held, not how volatile it is. The fund's value rises and falls with bitcoin, which has historically experienced very large price swings.

And the legitimacy of regulated funds is used as cover by fraudsters: fake "Bitcoin ETF" investment offers and platforms exist. Buy listed funds only through a genuine regulated broker, and see our scam checklist.

Related reading

• Pig Butchering Scams: The Long Con Now Run by AI
• AI Trading Bots: What They Can Do, and Why Regulators Keep Warning
• Ten Red Flags of a Financial Scam, From Forex to Crypto to AI
• Tokenised Treasuries: Wall Street's Quiet Move Onto the Blockchain
• Robo-Advisors Explained: What 0.25% a Year Actually Buys You

Sources

• "BlackRock's IBIT captures $479M as Bitcoin ETFs extend streak," Bitcoin.com News — news.bitcoin.com
• "BlackRock's IBIT leads Bitcoin ETFs to their biggest day since January," Coinpaprika — coinpaprika.com
• "Bitcoin ETF flows 2026 analysis," Intellectia — intellectia.ai
• "BlackRock IBIT sees $214M outflow as redemption streak hits $4.4B," Investing.com — investing.com

A Stablecoin Is a Promise That One Token Equals One Dollar. Now There Is a Law Behind It.

Stablecoins are crypto tokens pegged to a currency such as the US dollar. USDT and USDC hold about four-fifths of the market, and the GENIUS Act sets the first US federal rules, with one-to-one reserves and redemption rights.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how stablecoins work. It is general information, not financial or investment advice.

Bitcoin's price can swing by thousands of dollars in a day. That is useless for paying rent. Stablecoins exist to solve that problem — and they have quietly become one of the most important parts of the crypto system.

What a Stablecoin Is

A stablecoin is a crypto token designed to hold a fixed value, usually one US dollar. The issuing company says it holds reserves — cash, short-term government debt or similar assets — so that every token can be redeemed for a dollar.

In practice, stablecoins are used to move dollar value quickly between exchanges and wallets, to park money inside the crypto system without selling back to a bank, and increasingly for payments and cross-border transfers.

The Two Giants

Two tokens dominate:

• USDT, issued by Tether — around $189 billion in Q1 2026.
• USDC, issued by Circle — around $77 billion in Q1 2026.

By mid-April 2026, USDT made up roughly 58 percent of dollar-backed stablecoin supply, and USDT and USDC together accounted for about four-fifths of the whole market.

The balance is shifting. Tether's USDT supply contracted by about $3 billion in Q1 2026 — its first quarterly decline since 2022 — while Circle's USDC added about $2 billion to reach $78 billion, driven by institutional demand for regulated assets.

The GENIUS Act

The GENIUS Act, enacted in July 2025, gives the United States its first federal rulebook for payment stablecoins. It requires:

• Permitted issuers — only approved entities may issue payment stablecoins.
• Liquid one-to-one reserves backing every token.
• Disclosures about those reserves.
• Redemption procedures, so holders can convert back to dollars.
• Financial-crime controls.

The law takes effect on the earlier of 18 months after enactment — pointing to around January 2027 — or 120 days after final regulations are issued.

Why USDT and USDC Face It Differently

USDC, from a US-based issuer, appears positioned to move into the regulated framework, subject to approvals and final rules.

USDT's position is more complex because it is issued abroad. Tether is pursuing compliance while also using a separate US-focused token.

The Risks That Remain

Reserve risk. A stablecoin is only as good as the assets behind it and the honesty of the reporting. The reserve and disclosure rules exist because this has been a real concern.

Run risk. If many holders try to redeem at once and reserves are not liquid, a token can lose its peg.

Issuer risk. A stablecoin is a claim on a private company, not a bank deposit with government insurance.

Scam risk. Fraudsters frequently ask victims to pay in USDT precisely because it is stable, fast and hard to reverse — a pattern described in our guide to pig butchering scams.

Stablecoins Versus Central Bank Digital Money

A stablecoin is a private company's digital dollar. A central bank digital currency would be issued by the central bank itself. The two are competing visions of digital money, compared in our report on CBDCs.

Related reading

• MiCA Deadline Passed: About 80% of EU Crypto Firms Did Not Get a Licence
• What Moves Exchange Rates, and Why the Fed's 16 September Decision Matters
• Is the AI Stock Boom a Bubble? The Numbers on Both Sides
• Central Bank Digital Currencies: 134 Countries Exploring, 3 Fully Launched

Sources

• "What is the GENIUS Act? US stablecoin law explained for 2026," Eco — eco.com
• "2026 stablecoin laws: GENIUS Act rules for USDT and USDC," Stablecoin Laws — stablecoinlaws.org
• "GENIUS Act stablecoin rules 2026: what USDC and USDT holders face," Phemex Academy — phemex.com
• "GENIUS Act stablecoin compliance (July 2026)," Decentralfeed — decentralfeed.com

The Friend Who Taught You to Trade Crypto Was Never Real

Pig butchering scams build a fake friendship or romance, then move the victim onto a fake crypto platform showing invented profits. Crypto investment fraud cost Americans $7.2 billion in 2025, and AI now automates the grooming and fakes live video calls.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

The name is deliberately ugly, and it describes the method exactly: fatten the victim with fake gains, then "butcher" them for everything they have.

It is now the costliest fraud category the FBI tracks.

How It Works, Step by Step

1. The contact. A "wrong number" text, a dating app match, or a friendly message on social media. Nothing about money.

2. The relationship. Weeks or months of daily conversation. The scammer becomes a friend, a confidant or a romantic interest. They are successful, patient and attentive.

3. The hint. Eventually they mention how well they are doing from crypto trading, often with an "uncle" or "mentor" who has special insight.

4. The platform. They offer to help. They direct the victim to an "exclusive" trading platform. It is not a real exchange — investigators call these mirror dashboards, sites that look and function like a genuine trading interface.

5. The fattening. The victim deposits a small amount. The dashboard shows impressive profits. Sometimes a small withdrawal is allowed, to prove it works.

6. The butchering. Confident, the victim deposits much more — savings, retirement funds, borrowed money. When they try to withdraw, they are told to pay a "tax" or "fee" first. The money never comes out.

Why the Losses Are So Large

The victim believes they are making their own investment decisions on a legitimate platform. They are not being asked to send money to a person; they are "investing".

That is why losses are often far larger than in a conventional romance scam: victims keep adding money because the dashboard tells them their investment is growing.

In its 2025 report, the FBI's IC3 recorded $8.6 billion in investment-fraud losses, with $7.2 billion in cryptocurrency investment fraud — the category dominated by pig butchering. Wider figures are in our report on the IC3 data.

What AI Changed in 2026

Automated grooming. Criminal syndicates now use AI agents to run the relationship stage. One operation can maintain thousands of personalised conversations simultaneously, in any language, at any hour.

Live deepfake video. A scammer who once avoided video calls can now appear on a live call as an entirely fabricated person. "I saw them on video" is no longer proof anyone is real. See our report on deepfake fraud.

The Warning Signs

• A stranger who becomes a close friend or partner without ever meeting in person.
• Unprompted talk about investment success.
• A specific platform or app you must use, not a well-known exchange.
• Returns that are consistently high and never go down.
• Fees or taxes demanded before you can withdraw.
• Pressure to keep it secret from family or your bank.

A Simple Test

Before depositing anything, look up the platform independently — not through a link they sent. Check whether it is licensed by a real regulator, using the steps in our broker-checking guide. Then try to withdraw a meaningful amount early. A scam platform will find a reason you cannot.

If You Have Been Caught

• Stop paying. The withdrawal fee is part of the scam, and more will follow.
• Stop talking to the "friend"; they will try to keep you engaged.
• Save every message, address and transaction record.
• Report to police and your bank — and to IC3 if you are in the US.
• Ignore recovery offers that ask for payment upfront. Victims are often targeted a second time.

Victims are not foolish. These operations are industrial, patient and psychologically sophisticated. Shame is what keeps people sending money; talking to someone early is what stops it.

Related reading

• Ten Red Flags of a Financial Scam, From Forex to Crypto to AI
• AI Trading Bots: What They Can Do, and Why Regulators Keep Warning
• How to Check a Forex Broker Is Really Regulated, and Spot a Clone Firm
• FBI: $11.4 Billion Lost to Crypto Scams in 2025, Over Half of All Fraud

Sources

• "Pig butchering scams 2026: how they work and how to escape," Petronella Cybersecurity — petronellatech.com
• "2026's pig-butchering reckoning: how AI is supercharging crypto's deadliest scam," Crypto Impact Hub — cryptoimpacthub.com
• "2025 IC3 annual report," FBI Internet Crime Complaint Center — ic3.gov

More Than Half of All Money Lost to Online Fraud in America Now Involves Crypto

The FBI's 2025 Internet Crime Report recorded $20.9 billion in losses, of which $11.37 billion involved cryptocurrency — up 22%. Crypto investment fraud alone cost $7.2 billion, and people over 60 lost $4.4 billion.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

The FBI's Internet Crime Complaint Center (IC3) publishes the largest official record of online fraud reported by the public. Its 2025 report makes one trend impossible to miss.

The Headline Numbers

• $20.877 billion in total reported losses across all internet crime.
• $11.366 billion of that described as cryptocurrency-related — more than half.
• A 22 percent increase in crypto-related losses on the previous year.
• 181,565 crypto-related complaints, out of roughly 1 million in total.

These are only reported losses. Many victims never file a complaint, so the true figure is higher.

Investment Fraud Is the Biggest Category

Cryptocurrency investment fraud was the single largest source of financial loss to Americans in 2025, at $7.2 billion.

That category is dominated by long-con schemes in which the victim is groomed over weeks, then persuaded to "invest" on a fake platform that displays invented profits — explained in our guide to pig butchering scams.

Older People Lost the Most

Americans aged 60 and over filed 44,555 crypto-related complaints and reported $4.432 billion in crypto losses — more than any other age group.

Older adults are targeted deliberately: they are more likely to have savings, retirement funds and home equity, and less likely to be familiar with how crypto transactions work or how hard they are to reverse.

Why Crypto Suits Fraudsters

Irreversibility. A bank transfer can sometimes be recalled. A crypto transaction, once confirmed, generally cannot.

Speed across borders. Funds can move internationally in minutes, often beyond the reach of the victim's local police.

Unfamiliarity. Victims who do not understand wallets and exchanges are easier to direct to fake platforms.

Legitimacy by association. Real crypto markets, regulated Bitcoin ETFs and genuine price rallies give fake "opportunities" a believable backdrop.

AI Is Now Its Own Category

2025 was also the first year the FBI logged a dedicated AI-related crime category, with more than 22,000 complaints and roughly $893 million in losses. Voice cloning and deepfake video are increasingly part of investment scams — see our report on deepfake fraud.

What Is Working

The FBI's Operation Level Up, which targets crypto investment scams, has notified more than 8,000 victims — many of whom did not yet realise they were being defrauded — and helped prevent more than $500 million in losses, including $225.9 million in 2025 alone.

Early intervention matters because the largest losses come from victims who keep sending money after the first deposit.

If It Is Happening to You or Someone You Know

• Stop sending money immediately, including any "fee" or "tax" demanded to release funds.
• Keep everything — wallet addresses, transaction IDs, messages and website links.
• Report it to the police and, in the US, to IC3.
• Beware of "recovery" services that promise to get the money back for an upfront fee. They are frequently a second scam aimed at the same victim.

Related reading

• Pig Butchering Scams: The Long Con Now Run by AI
• AI Trading Bots: What They Can Do, and Why Regulators Keep Warning
• Ten Red Flags of a Financial Scam, From Forex to Crypto to AI
• Deepfake Fraud: Three Seconds of Audio Can Clone a Voice
• How Banks Use AI to Catch Fraud Before the Money Leaves

Sources

• "2025 IC3 annual report," FBI Internet Crime Complaint Center — ic3.gov
• "FBI report finds crypto scams accounted for over $11B in losses in 2025," Fox Business — foxbusiness.com
• "FBI IC3 report: half of 2025 US fraud losses were linked to cryptocurrency scams," ASIS Security Management — asisonline.org
• "IC3 report reveals surge in cryptocurrency investment scams," Forbes — forbes.com
• "Deepfake statistics 2026," StationX — stationx.net

The Funded-Trader Business Makes Most of Its Money From People Who Fail

Retail prop trading firms sold an estimated 12 million challenges in 2026. Between 5% and 14% end in a funded account and only about 7% of buyers ever receive a payout, while payout disputes became the top complaint.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how the prop firm industry works. It is general information, not financial or investment advice.

"Trade our capital, keep up to 90 percent of the profit." It is one of the most heavily advertised offers in online trading. The industry's own statistics explain how the model actually works.

How a Challenge Works

You pay a fee to take an evaluation — the "challenge". You trade a demo or evaluation account under strict rules: a profit target, a maximum daily loss, a maximum total drawdown, and often consistency requirements.

Pass one or more phases and you receive a "funded" account. Profits you make there are shared with you. Break a rule at any stage and the account ends; the fee is gone.

The Numbers

• 5 to 14 percent of purchased challenges end in a funded account.
• About 7 percent of all challenge buyers ever receive a payout.
• Challenge pass rates in 2026 remain between 5 and 10 percent.

The industry is large and growing. Retail prop trading generates an estimated $850 million in annual revenue in 2026, up 45 percent year on year, across about 2.1 million active funded traders and 12 million challenge purchases.

The top five firms — FTMO, FundedNext, The 5%ers, Apex Trader Funding and TopStep — control an estimated 62 percent of the market by trader acquisition.

In 2025, prop firms paid out approximately $325 million to traders.

Why the Model Works for the Firm

Set revenue against payouts and the structure becomes clear. Most revenue comes from challenge fees, and most challenge buyers fail. The firm does not need traders to succeed to be profitable; it needs a steady stream of new challenge purchases.

That is not automatically dishonest — many firms do pay successful traders. But it means the headline promise of trading "their capital" describes a small minority of customers.

The Complaints

In 2025, payout-denial disputes became the top complaint category in the sector. Payouts depend on passing multi-phase evaluations and complying with consistency clauses, and disagreements over whether a rule was broken are common.

Before buying a challenge, read the rules that govern payouts — not just the profit split.

The Regulatory Grey Zone

US regulators view some simulated-trading payout models as falling within their reach. The industry's response has been to market itself as evaluation services rather than investment products, to separate trader losses from company equity, and increasingly to move from purely simulated challenges to live trading.

For customers, that means consumer protections that apply to brokers — such as those described in our guide to broker regulation — often do not apply to prop firms at all.

Before You Buy a Challenge

• Treat the fee as spent. Statistically, that is the likely outcome.
• Read the drawdown and consistency rules in full, especially how daily loss is calculated.
• Search for payout complaints about the specific firm.
• Be wary of firms that sell endless discounted retries — the retry is the product.
• Remember the underlying odds. Most retail traders lose money, and a challenge does not change the market.

Related reading

• MiCA Deadline Passed: About 80% of EU Crypto Firms Did Not Get a Licence
• What Forex Trading Is, and Why 71% of Retail Accounts Lose Money
• How a Central Bank Rate Decision Reaches Your Loan and Savings Account
• Leverage Explained: How 30:1 Turns a Small Move Into a Wiped Account

Sources

• "Prop trading statistics 2026: pass rates and market data," Track360 — track360.io
• "Prop firm statistics 2026: pass rates, payouts and industry data," Atmos Funded — atmosfunded.com
• "What percentage of traders pass prop firm challenges? (2026)," Pura Vida Edge — puravidaedge.com
• "Why traders fail prop firm challenges," Velotrade — velotrade.com

Why a Currency Rises or Falls, Explained Through This Week's Fed Meeting

Interest rates, inflation and risk appetite drive currencies. With US rates at 3.50–3.75% and markets pricing a possible hike at the Fed's 16 September meeting, here is how those forces reach the dollar and everything priced against it.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how exchange rates work. It is general information, not financial or investment advice.

Exchange rates look random from the outside. They are not. A handful of forces do most of the work, and this week offers an unusually clear example of the most important one.

Force One: Interest Rates

Money flows toward higher returns. When one country's central bank raises interest rates, holding that currency pays more, so demand for it tends to rise.

The US federal funds target range is 3.50 to 3.75 percent. The Fed has held it there since December 2025, most recently on 29 July 2026.

The next decision comes on Wednesday 16 September 2026, at the end of a two-day meeting that also publishes updated economic projections.

What Markets Expect

Ahead of the meeting, pricing moved toward a rate rise. One widely followed market-implied measure put the probability of a 0.25-point increase — to 3.75–4.00 percent — at about 85 percent.

That shift followed a hawkish speech by Fed Chair Warsh at Jackson Hole on 28 August and a solid August jobs report.

Probabilities like this are snapshots of market bets, not forecasts, and they change quickly. The counter-argument is visible in the data: headline US consumer inflation has fallen for two months running, to 3.4 percent in July, which argues for patience.

Force Two: Inflation

Inflation erodes what a currency buys. Over long periods, currencies of high-inflation economies tend to weaken against those of low-inflation economies.

In the short run, though, inflation often moves currencies through interest rates: high inflation makes central banks more likely to raise rates, which can strengthen the currency first.

Globally, the IMF projects headline inflation rising to 4.7 percent in 2026, driven mainly by energy and food prices.

Force Three: Risk Appetite

When investors are nervous, money tends to move into assets seen as safe. The US dollar often benefits in those moments, as does gold.

When confidence returns, money flows back toward higher-yielding and riskier currencies and assets.

Why a Fed Decision Moves Everything

The dollar sits at the centre of global trade, debt and commodity pricing. A change in US rates affects:

• Other currencies, which move against the dollar.
• Borrowing costs worldwide, because a large share of international debt is in dollars.
• Commodity prices, many of which are quoted in dollars.
• Risk assets such as crypto, which have tended to react to shifts in rate expectations.

How a rate change reaches household loans and savings is covered separately in our guide to interest rates and your money.

The Honest Limit

Knowing what moves currencies does not make them predictable. Expectations are usually priced in before a decision; what moves the market is the surprise relative to those expectations. That is one reason most retail currency traders lose money.

Related reading

• How a Central Bank Rate Decision Reaches Your Loan and Savings Account
• Central Bank Digital Currencies: 134 Countries Exploring, 3 Fully Launched
• Stablecoins Explained: USDT, USDC and America's GENIUS Act
• What Forex Trading Is, and Why 71% of Retail Accounts Lose Money

Sources

• "Next Fed interest rate decision: 16 September 2026 preview," Cambridge Currencies — cambridgecurrencies.com
• "Fed rate decision: Wednesday, September 16, 2026," FedRateCalc — fedratecalc.com
• "Fed rate probability: FOMC meeting odds," Central Bank Watch — centralbank.watch
• "FOMC minutes, July 28–29, 2026," Board of Governors of the Federal Reserve System — federalreserve.gov
• "World Economic Outlook," International Monetary Fund — imf.org

The Licence Number on the Website Could Belong to Someone Else

Scam brokers copy the name, registration number and address of genuinely authorised firms. Checking the regulator's own register, and contacting the firm only through the details listed there, is the step that catches them.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article is general information, not financial or legal advice.

A professional website, a registration number in the footer and a London address look convincing. None of it proves the firm is who it says it is.

What a Clone Firm Is

The UK's Financial Conduct Authority describes the tactic directly: fake trading and brokerage firms use the name, firm registration number (FRN) and address of firms and individuals that genuinely are FCA-authorised. This is called a clone firm.

The registration number is real. The address is real. The company on the register is real. The website, phone number and bank account are not theirs.

The FCA regularly publishes warnings naming such clones, including firms trading under names such as "Forex Premium" and "Forex-max", flagged as clones of authorised firms.

The Five-Minute Check

1. Search the regulator's own register. In the UK, that is the FCA Financial Services Register, via its Firm Checker. Do not use a link from the broker's website — type the regulator's address yourself.
2. Confirm the permissions. Being on the register is not enough. The firm must be authorised for the specific service it is offering you.
3. Compare the contact details. Check the website address, phone number and email on the register against the ones you were given. A mismatch is the signature of a clone.
4. Contact the firm only through the register's details. The FCA advises that if a financial business contacts you unexpectedly, you reply using the contact details listed on the Firm Checker — not the ones in the message.
5. Search the warning list. The FCA Warning List names unauthorised firms and individuals that are not allowed to operate in the UK.

Other major regulators — including Cyprus's CySEC and Australia's ASIC — run their own public registers. The same principle applies everywhere: verify on the regulator's site, not the broker's.

What You Lose With an Unauthorised Firm

According to the FCA, if you deal with an unauthorised firm:

• You cannot complain to the Financial Ombudsman Service.
• You are not protected by the Financial Services Compensation Scheme if the firm goes out of business.

In practice, money sent to an unauthorised or clone firm is very rarely recovered.

Warning Signs Before You Even Check

• Unexpected contact — a call, message or social media approach you did not initiate.
• Pressure to decide quickly, or a "limited-time" bonus.
• Leverage far above regulated limits — covered in our leverage guide.
• Promises of guaranteed or consistent returns.
• Requests to pay by cryptocurrency or to an individual's bank account.
• Difficulty withdrawing, or new fees demanded before a withdrawal is released.

If You Think You Have Been Targeted

The FCA asks people to report suspected scams, and says it looks into every report — which can help protect others. Stop sending money, keep every record of the conversation and payments, and contact your bank immediately.

For a wider list of red flags across forex, crypto and AI-themed schemes, see our scam checklist.

Related reading

• Ten Red Flags of a Financial Scam, From Forex to Crypto to AI
• Pig Butchering Scams: The Long Con Now Run by AI
• Deepfake Fraud: Three Seconds of Audio Can Clone a Voice
• AI Trading Bots: What They Can Do, and Why Regulators Keep Warning

Sources

• "Forex trading scams," Financial Conduct Authority — fca.org.uk
• "FCA Warning List of unauthorised firms," Financial Conduct Authority — fca.org.uk
• "Financial Services Register," Financial Conduct Authority — fca.org.uk
• "Forex Premium (clone of FCA authorised firm)," Financial Conduct Authority — fca.org.uk

Leverage Multiplies Everything. That Is the Whole Problem.

EU rules cap retail leverage at 30:1 on major currency pairs and 2:1 on crypto, force positions closed at 50% of required margin, and guarantee you cannot lose more than your deposit. What that means in real numbers.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how leverage works. It is general information, not financial or investment advice.

Leverage is the reason forex adverts can promise that a small deposit controls a large position. It is also the main reason around 71 percent of retail accounts lose money.

What Leverage Is

Leverage lets you open a position larger than the money in your account. At 30:1, a deposit of $1,000 controls a position worth $30,000. The deposit you put up is called margin.

The broker is not being generous. Profits and losses are calculated on the full $30,000, not on your $1,000.

The Arithmetic, Plainly

Take that $30,000 position on $1,000 of margin.

• If the price moves 1 percent in your favour, you make $300 — a 30 percent return on your deposit.
• If it moves 1 percent against you, you lose $300 — 30 percent of your deposit.
• If it moves about 3.3 percent against you, the loss is $1,000 — your entire deposit.

Major currencies can move that much in a single day of news. That is why leverage is the single most dangerous feature of retail trading.

The EU Limits

ESMA, the European Securities and Markets Authority, set leverage limits for retail clients that vary with how volatile the asset is:

• 30:1 — major currency pairs.
• 20:1 — non-major currency pairs, gold and major stock indices.
• 10:1 — commodities other than gold, and non-major indices.
• 5:1 — individual shares.
• 2:1 — cryptocurrencies.

These measures took effect on 1 August 2018. They were introduced as temporary and remain in place.

The Margin Close-Out Rule

ESMA also requires brokers to close a retail client's positions when their account falls to 50 percent of the minimum required margin.

In the example above, the minimum margin is $1,000. Once losses bring the account's equity down to around $500, the broker closes the position automatically — roughly a 1.7 percent adverse move at 30:1. The trade is over whether or not the trader agrees.

This is what traders mean by being "stopped out" or hitting a margin call.

Negative Balance Protection

ESMA mandates negative balance protection per account, so a retail trader cannot lose more than the money in their account.

Before these rules, a violent market move could leave a trader owing the broker money beyond their deposit. In a sudden currency shock, prices can gap past any stop — negative balance protection is what prevents a bad day from becoming a debt.

Why Offshore Leverage Is a Red Flag

Brokers outside these rules advertise leverage of 500:1 or more. At 500:1, a 0.2 percent move erases the deposit entirely.

High advertised leverage is not a feature. It usually signals a firm operating outside the regulations built to protect retail clients — see how to check a broker's regulation.

The Practical Takeaways

• Calculate the percentage move that would wipe out your deposit before opening a trade.
• Lower leverage than the maximum is always available. Using it is a choice.
• Choose brokers that provide negative balance protection.
• Treat any offer of very high leverage as a warning, not an opportunity.

Related reading

• Prop Firm Challenges: Only About 7% of Buyers Ever Get Paid
• Tokenised Treasuries: Wall Street's Quiet Move Onto the Blockchain
• How a Central Bank Rate Decision Reaches Your Loan and Savings Account
• What Forex Trading Is, and Why 71% of Retail Accounts Lose Money

Sources

• "ESMA agrees to prohibit binary options and restrict CFDs to protect retail investors," European Securities and Markets Authority — esma.europa.eu
• "ESMA to renew restriction on CFDs for a further three months," European Securities and Markets Authority — esma.europa.eu
• "CFD retail broker leverage limits by regulator," Liquidity Finder — liquidityfinder.com
• "Why do I have trading restrictions on certain products? Retail vs professional client accounts," Saxo — help.saxo

Forex Brokers Must Tell You How Many Customers Lose. The Answer Is Most of Them.

Mandatory disclosures from 49 EU and UK-regulated brokers show an average of 71% of retail forex and CFD accounts lose money, with individual brokers ranging from 51% to 81%. How the market works and why the odds look like that.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

This article explains how forex trading works. It is general information, not financial or investment advice.

Forex adverts show beaches, laptops and freedom. The small line at the bottom of every regulated advert in Europe and the UK tells a different story, and it is required by law.

What Forex Trading Actually Is

Forex — foreign exchange — is the buying of one currency with another. When a trader "buys EUR/USD", they are betting the euro will rise against the US dollar.

Banks, companies and governments use this market to pay for imports, hedge risk and manage reserves. Retail traders mostly access it through CFDs — contracts for difference — which let them bet on price movements without owning the currency, usually with borrowed money.

The Number Brokers Must Publish

In the European Union and the UK, CFD promotions must carry a standardised warning stating the percentage of that broker's own retail accounts that lose money. It has to be the firm's current figure, displayed prominently — including in social media posts and banner ads.

Across mandatory disclosures from 49 EU and FCA-regulated brokers:

• Average: 71.0 percent of retail accounts lose money.
• Median: 71.7 percent.
• Range: 51 percent to 81 percent, depending on the broker.

In the UK specifically, the April 2026 reading across 14 brokers was a 69.9 percent mean and 71.0 percent median. Named examples include 68 percent at CMC Markets UK, 68 percent at IG UK and 71 percent at IG International.

The original ESMA standard warning cited a range of 74 to 89 percent. Current figures are somewhat lower, but the picture is the same.

Why Most Retail Traders Lose

Leverage. Borrowed money multiplies gains and losses equally. A small move against a leveraged position can erase the deposit — explained in detail in our guide to leverage.

Costs. Every trade pays a spread, and positions held overnight pay financing. Frequent trading means paying those costs repeatedly, whether the trade wins or not.

The competition. A retail trader is on the other side of banks, hedge funds and algorithmic firms with better data, faster execution and far larger resources.

Behaviour. Holding losing trades in the hope they recover, and closing winning trades early to lock in a gain, is a pattern that reliably loses money over time.

Where the Industry Is Heading

Even parts of the trading industry now warn retail customers away from CFD brokers. Some firms have leaned more heavily on "educational" labels as marketing rules have tightened.

Proprietary trading firms, which sell evaluation challenges rather than brokerage accounts, have grown as an alternative — with their own problems, covered in our report on prop firms.

If You Are Still Considering It

• Find the broker's own loss percentage before opening an account. A regulated broker must show it.
• Check the broker is genuinely regulated — here is how.
• Only use money you can afford to lose completely. That is not a figure of speech; it is what the statistics describe.
• Be deeply sceptical of anyone selling signals, courses or bots that promise consistent profit. See what regulators say about AI trading bots.

Related reading

• AI Trading Bots: What They Can Do, and Why Regulators Keep Warning
• How a Central Bank Rate Decision Reaches Your Loan and Savings Account
• Prop Firm Challenges: Only About 7% of Buyers Ever Get Paid
• What Moves Exchange Rates, and Why the Fed's 16 September Decision Matters

Sources

• "71.0% of retail traders lose money — 2026 data (49 brokers)," BrokerRank — brokerrank.net
• "UK CFD trading statistics 2026 — 14 brokers compared," The Investors Centre — theinvestorscentre.co.uk
• "Forex broker marketing compliance 2026: rules and disclosure," Track360 — track360.io
• "Prop firm E8 Markets warns retail traders off CFD brokers," Finance Magnates — financemagnates.com

AI Can Help You Understand Your Health. It Should Not Diagnose It.

More than 40 million people a day ask ChatGPT health questions, and misuse of AI chatbots tops ECRI's 2026 list of health technology hazards. A study found unsafe answers in 5% to 13% of cases across models.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

More than 40 million people a day turn to ChatGPT for health information. For people without easy access to a doctor, a patient explanation at midnight feels like a gift.

It can be. It can also be dangerous, and 2026 produced strong evidence about exactly where the line sits.

This article is general information, not medical advice. If you think you may have an emergency, contact a doctor or emergency services immediately.

The Warnings

ECRI, the patient safety organisation, ranked misuse of AI chatbots as the top health technology hazard for 2026, citing rapid adoption, lack of regulatory oversight, and mounting evidence that chatbots generate unsafe or misleading medical guidance.

A central concern: general chatbots are not regulated as medical devices and not validated for healthcare purposes, yet they are used by patients, clinicians and health staff.

A University of Oxford study published in February 2026 warned that asking AI about symptoms can lead to wrong diagnoses and a failure to recognise when urgent help is needed.

What the Numbers Showed

A study in npj Digital Medicine tested major models on medical questions posed as patients would ask them.

• Problematic responses ranged from 21.6 percent (Claude) to 43.2 percent (Llama).
• Unsafe responses ranged from 5 percent (Claude) to 13 percent (GPT-4o and Llama).

Even the best result means roughly one answer in twenty could be unsafe. No reasonable person would accept that failure rate from a doctor.

Why Chatbots Get Health Wrong

A chatbot cannot examine you, order a test, or see how unwell you look. It works only from what you type — and people describe symptoms incompletely, especially when frightened.

It also shares the general weakness described in our guide to AI errors: it produces fluent, confident text whether or not it is correct.

Where AI Genuinely Helps

• Understanding a diagnosis you already have — in plain language, at your own pace.
• Preparing questions to ask your doctor at your next appointment.
• Explaining medical words in a test report or discharge letter.
• General wellbeing information — diet, sleep, exercise — checked against reliable sources.

Where It Should Not Be Used

• Deciding whether symptoms are an emergency. Chest pain, difficulty breathing, sudden weakness, severe bleeding, or signs of stroke need immediate medical help — not a chatbot.
• Changing or stopping medication.
• Diagnosing a child, a pregnancy complication, or mental health crisis.
• Replacing a clinician's judgement about treatment.

Five Rules for Safer Use

1. Use AI to understand, not to decide.
2. Tell it to advise seeing a doctor when appropriate, and take that advice.
3. Check important information against a reliable health authority.
4. Do not paste identifying health records into free tools — see our privacy guide.
5. When in doubt, contact a real clinician.

The World Health Organization's guidance on large multi-modal models in health includes specific material on use by patients and the public, and is the right starting point for anyone designing health services around these tools.

The Better Route in Bangladesh

The real answer to limited access is not a general chatbot. It is connecting people to qualified care faster.

That is what Bangladesh's healthtech and telemedicine sector is building, alongside a health budget that has nearly doubled and the public health record WHO recognised this month. A telemedicine consultation with a licensed doctor, even a short one, is a far safer use of a phone than a symptom checker with no one accountable behind it.

Related reading

• Is Your Data Safe With AI Chatbots? How to Opt Out of Training
• Vibe Coding: Why AI-Written Code Fails Security Tests
• Why AI Makes Things Up, and a Five-Step Way to Check It
• How Well Does AI Understand Bangla? A Practical Guide
• AI Detectors Flagged 61% of Non-Native English Essays as AI

Sources

• "Misuse of AI chatbots tops annual list of health technology hazards," ECRI — ecri.org
• "New study warns of risks in AI chatbots giving medical advice," University of Oxford — ox.ac.uk
• "Large language models provide unsafe answers to patient-posed medical questions," npj Digital Medicine — nature.com
• "Misuse of AI chatbots in health care tops 2026 Health Tech Hazard Report," Association of Health Care Journalists — healthjournalism.org

Most Small Businesses Get AI's Value From Three Boring Jobs

Small businesses using AI report saving about 5.6 hours a week, with 71% reporting higher productivity. Marketing content, customer messages and admin are where most start. A practical first-month plan.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

Small business owners are told AI will transform everything. What most actually need to know is simpler: where to start on Monday, and whether it is worth the time.

How Many Small Businesses Actually Use AI

It depends on how you count, and the gap between surveys is worth understanding.

Intuit found 77 percent of US small and midsize businesses use AI regularly, up from 48 percent in 2024. By stricter measures of use in core operations, the Federal Reserve found 46 percent and the Census Bureau 17 to 20 percent.

The honest reading: most small businesses have tried AI tools; far fewer have built them into how the business runs.

The Three Uses That Lead

• Marketing content creation — 68 percent of small businesses using AI.
• Customer communication — 52 percent.
• Administrative tasks — 47 percent.

None of these is glamorous. All of them are tasks the owner was doing late at night.

What It Returns

Among small businesses using AI, 71 percent reported higher productivity, 39 percent better quality and 31 percent higher sales.

The average small business saves about 5.6 hours a week; owners and managers save more than 7 hours. For a one-person business, that is close to a working day returned every week.

The Real Barrier Is Not Cost

Across EU, OECD, UK and G7 surveys, 50 to 71 percent of businesses that have not adopted AI cite lack of expertise as the main reason — ahead of cost, regulation and data privacy.

That is good news. Expertise can be learned quickly for the uses above. Paid plans for small-business AI agent platforms typically run $29 to $99 a month, and the median small business uses around five AI tools.

A First-Month Plan

Week 1 — Marketing. Use a general AI assistant to draft this week's social media posts and product descriptions. Give it examples of posts that worked before. Edit everything before posting. Learn the basics in our prompting guide.

Week 2 — Customer messages. Write your ten most common customer questions and your best answers. Use AI to draft replies in that style. If you sell through chat, note that Meta's business agent is already used by over a million businesses on WhatsApp, Messenger and Instagram.

Week 3 — Admin. Use AI to summarise supplier emails, draft invoices and quotes, and turn messy notes into organised lists.

Week 4 — Measure. Count the hours saved. Keep what saved time; drop what did not. Only then consider something more ambitious, such as a simple AI agent.

Three Mistakes to Avoid

• Publishing unchecked output. A wrong price or invented claim costs trust. See how to check AI answers.
• Pasting customer data into free tools. Read our privacy guide first.
• Buying five subscriptions in week one. Start with one tool and a clear task.

For Bangladesh's Small Businesses

Bangladesh has 2.8 million women-led SMEs and a huge base of small online sellers, many operating through Facebook, Instagram and WhatsApp as described in our e-commerce reporting. For a business where the owner answers every message personally, saving seven hours a week is the difference between staying small and being able to grow.

The tools are already on the phone in their hand. The skill to use them well is the part worth investing in.

Related reading

• Anthropic Has Overtaken OpenAI in Business AI Spending
• AI for Freelancers: The Work That Still Pays, and How to Move Into It
• What an AI Agent Actually Is, and How It Differs From a Chatbot
• How Much Electricity and Water Does AI Actually Use?

Sources

• "Small business AI adoption statistics for 2026," Capsule CRM — capsulecrm.com
• "SME AI adoption in 2026: what the data actually shows," Omago — omago.ai
• "The AI tools small businesses are using," Small Business & Entrepreneurship Council — sbecouncil.org
• "AI for small business 2026: adoption rates, tools, and implementation roadmap," Octopus Builds — octopusbuilds.com

A Freelancer's Playbook for the AI Market

Freelancers using AI for complex work earned 45% more year on year, and AI-augmented professional services grew 72%. A practical plan for freelancers to move from standardised tasks to higher-paying work.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

If you freelance, you have probably noticed two things at once: some kinds of gigs are drying up, and some clients are paying more than ever. Both are the same trend.

Our companion piece, Will AI take my job?, sets out the evidence. This one is the practical plan.

What the Market Is Paying For Now

Upwork's Future Workforce Index 2026 found freelancers doing more complex work with AI saw earnings rise 45 percent year on year, while AI-augmented professional services grew 72 percent in volume, with earnings also rising.

Meanwhile stock photography, template design, commodity copywriting and basic video editing are declining. The market has not stopped buying work. It has stopped paying people for output a tool now produces in seconds.

Step 1: Audit Your Current Gigs

List your last ten paid jobs. Mark each one:

• Standardised — the result looks broadly the same each time.
• Judgement-based — it depended on understanding a specific client, market or problem.

If most are standardised, that income is exposed. The goal is not to abandon it overnight, but to shift the mix steadily toward the second column.

Step 2: Use AI to Do the Standardised Part Faster

Refusing AI tools is not a strategy. Clients can see the market rate, and it has moved.

Use AI for first drafts, research summaries, variations, code scaffolding and formatting. Then spend your time on the part that decides quality: structure, accuracy, judgement and the client's actual goal. Good prompting is now a basic professional skill.

Step 3: Sell Outcomes, Not Hours

If a task that took you eight hours now takes two, billing hourly cuts your income by three-quarters. Price the result — the landing page that converts, the working feature, the finished report — not the time.

Step 4: Add a Domain

The fastest-growing category is domain expertise plus AI. A general writer competes with every AI model. A writer who understands garment export compliance, pharmaceutical regulation or South Asian logistics does not.

Bangladesh has deep, internationally relevant expertise in exactly such areas: apparel sourcing, generic pharmaceuticals, mobile financial services, climate adaptation. Freelancers who pair that knowledge with AI skills are selling something scarce.

Step 5: Become the Person Who Checks

AI output needs verification. It invents citations and numbers, and AI-written code fails security tests at high rates.

Clients increasingly need someone accountable for getting it right: reviewing AI code for security, fact-checking AI content, testing AI-built workflows. That is a service in its own right.

Step 6: Offer AI Implementation

Small businesses cite lack of expertise as the main reason they do not adopt AI — ahead of cost. That gap is a market.

Setting up a simple AI agent, connecting tools, writing a company's AI usage policy, or training a team are all services a skilled freelancer can package and sell.

Step 7: Protect Client Data

Pasting a client's confidential material into a free AI tool that trains on conversations can breach a contract. Read our privacy guide and state your data practices in your proposals. Serious clients notice.

The Bangladesh Opportunity

Bangladesh built one of the world's largest online freelance workforces by competing on price for standardised digital work. That model is under real pressure.

The replacement model is available and pays better: judgement, domain knowledge and verification, delivered with AI speed. The freelancers who make that move in the next two years will be in a far stronger position than those who wait.

Related reading

• Will AI Take My Job? What the 2026 Evidence Actually Shows
• AI for Small Business: Where to Start, and What It Actually Saves
• MCP Explained: The Protocol That Connects AI to Your Tools
• GPT-6 Astra and What It Means for Bangladesh

Sources

• "Upwork's Future Workforce Index 2026," Upwork via GlobeNewswire — globenewswire.com
• "The 2026 AI job disruption report," AI Magicx — aimagicx.com
• "SME AI adoption in 2026: what the data actually shows," Omago — omago.ai
• "Top 20+ predictions from experts on AI job loss," AIMultiple — aimultiple.com

AI's Electricity Bill Is Real. So Is the Efficiency Curve.

Data centre electricity use grew 17% in 2025 while AI-focused data centres rose 50%. The IEA projects total data centre demand roughly doubling from 485 TWh to 950 TWh by 2030, about 3% of global electricity.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

"How much energy does AI use?" is one of the most searched questions about the technology. The answers online range from dismissive to apocalyptic. The International Energy Agency's figures are the most useful place to start.

The Electricity Numbers

Global data centre electricity demand grew 17 percent in 2025. Electricity consumption from AI-focused data centres surged 50 percent in the same year.

The IEA projects total data centre consumption roughly doubling from 485 TWh in 2025 to about 950 TWh in 2030 — around 3 percent of global electricity demand. Power use from AI-focused data centres is expected to triple to about 465 TWh, approaching the consumption of conventional data centres.

Three percent of world electricity is significant. It is also smaller than the share many alarming headlines imply.

The Water Question

Data centres use water in two ways. Directly, to absorb and remove the heat servers generate. Indirectly, through the water consumed in producing the electricity they run on.

The direct use is concentrated and locally visible, which is why it draws protest in water-stressed regions. The good news is technical: liquid cooling — immersion or direct-to-chip — reduces direct water use by 70 to 90 percent.

That does not solve the larger issue. Even with efficient cooling, the underlying electricity demand remains the dominant environmental concern.

The Part That Is Improving

The IEA notes that power consumption per AI task is declining rapidly, with efficiency improving at a rate it describes as unprecedented in energy history.

Both things are true at once. Each individual AI request is getting much cheaper in energy terms. Total demand is still rising, because use is growing even faster than efficiency improves.

That pattern is familiar from lighting, computing and transport: cheaper per unit usually means more units.

Why the Grid Is Now the Bottleneck

AI capacity is now measured in gigawatts rather than chips. As covered in our report on Australia's 2-gigawatt AI buildout, the question a country faces is whether it has the generation, transmission and cooling to run the hardware — not whether it can buy it.

Where that power comes from decides the climate impact. That is why the shift of investment into solar-plus-storage matters to AI as much as to households.

What It Means for Bangladesh

Bangladesh's data centre sector is growing, with Tier-IV facilities under development. Every one of them runs into the country's longest-standing industrial constraint: reliable, affordable electricity.

For a hot, humid climate, cooling efficiency matters even more than elsewhere. And for an energy importer already facing higher energy prices, siting AI capacity alongside renewable generation is an economic decision as much as an environmental one.

What You Can Do as a User

Individual chat queries are a small part of the total, and guilt about asking a question is misplaced. The choices that matter are at scale:

• Use the smallest model that does the job. It is cheaper, faster and uses less energy.
• Avoid sending huge documents unnecessarily. See our context window guide.
• For routine work, consider a local model on hardware you already own — here is how.
• For businesses, ask providers where their power comes from. Procurement pressure is what moves data centre energy policy.

Related reading

• Is Your Data Safe With AI Chatbots? How to Opt Out of Training
• Will AI Take My Job? What the 2026 Evidence Actually Shows
• What an AI Agent Actually Is, and How It Differs From a Chatbot
• AI for Small Business: Where to Start, and What It Actually Saves

Sources

• "Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions," International Energy Agency — iea.org
• "Key questions on energy and AI: executive summary," International Energy Agency — iea.org
• "Data centers and water fact sheet," Environmental Law Institute — eli.org
• "Global energy demands within the AI regulatory landscape," Brookings — brookings.edu

Training Was Fair Use. Pirated Copies Were Not. The Copyright Map So Far.

Around 130 AI copyright lawsuits are being tracked. One court found training on books was fair use but keeping pirated copies was not, producing a $1.5 billion settlement worth about $3,000 per work.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

Artificial intelligence companies face a wave of litigation from authors, artists and media organisations over the unlicensed use of copyrighted work to train their models. Trackers now count around 130 AI copyright lawsuits across US and international courts.

This is not legal advice. It is a map of where the law stands, because both creators and everyday AI users are affected.

The Ruling That Drew the First Line

The most consequential decision so far split the question in two.

A court ruled that training AI on copyrighted books constituted fair use — but that storing pirated copies did not. The outcome was a $1.5 billion settlement, amounting to roughly $3,000 per work.

The distinction matters. The ruling did not say AI companies may use anything freely. It said the act of learning from a work can be lawful while the way the work was obtained can still be unlawful.

Outputs Are a Separate Question

On 12 March 2026, Judge Valerie Chen ruled that AI-generated outputs based on copyrighted training data may constitute infringement if they are "substantially similar" to protected works.

So even where training is permitted, an AI that reproduces a protected work closely enough can still infringe.

The Cases Still Open

The New York Times v. OpenAI. The Times argues ChatGPT can reproduce its articles nearly verbatim. The case was still ongoing as of April 2026, and its appeal sits in the Second Circuit.

Authors v. Meta. A class action alleges Meta trained its Llama models on pirated book datasets. Clearly pirated training material is considered the weakest fair-use position for AI companies.

The appeals courts. The Ninth Circuit is positioned to be the first appellate court to rule directly on training as fair use, with the Second Circuit close behind on the Times appeal and the consolidated Authors Guild action. Those rulings will shape the law far more than any single trial.

Who Owns What AI Makes?

This is the question most everyday users actually have.

In many legal systems, copyright protection depends on human authorship. Output produced purely by a machine, with little human creative input, may not be protectable at all — meaning you may not be able to stop others copying it.

The more substantial your own creative contribution — selecting, arranging, editing, combining — the stronger your claim. Rules differ between countries and are still developing, so treat important commercial work with care.

Practical Guidance

If you use AI to create content:

• Do not ask it to imitate a named living artist or reproduce a specific work.
• Check the provider's terms for commercial use — they differ.
• Add real creative work of your own if ownership matters to you.
• Label AI-generated images, as described in our image generation guide.

If you create original work:

• Keep dated records of your drafts and originals.
• Review the terms of platforms where you publish; some grant AI training rights.
• Follow collective actions in your field. Settlements in the book cases paid individual rights-holders.

Why It Matters Beyond America

Most of these cases are in US courts, but their outcomes set the terms on which global AI companies operate everywhere.

For Bangladesh's creative economy — animation studios, streaming content makers, filmmakers and designers — the question is both defensive and commercial. Protecting original work matters. So does knowing which AI tools can be used safely in paid client projects.

Related reading

• Startups Raised $510 Billion in Six Months, More Than All of Last Year
• Vibe Coding: Why AI-Written Code Fails Security Tests
• Is Your Data Safe With AI Chatbots? How to Opt Out of Training
• Anthropic Has Overtaken OpenAI in Business AI Spending

Sources

• "AI in litigation series: an update on AI copyright cases in 2026," Norton Rose Fulbright — nortonrosefulbright.com
• "AI copyright lawsuits (2026): 130 cases tracked," AI Lawsuit Tracker — ailawsuittracker.com
• "AI copyright lawsuits heat up: March 2026 court rulings and industry takeaways," Tech Daily Shot — techdailyshot.com
• "AI copyright training data 2026: status, timeline, risk," AI Vortex — aivortex.io

The AI Detector Is Least Accurate for Students Writing in a Second Language

A 2026 study found leading AI detectors only 69% and 61% accurate, falling to nearly zero on mixed human-AI text. Earlier research found 61% of essays by non-native English writers wrongly flagged as AI.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

Across the world, students are being told their essays were written by AI — by software. Many of those students wrote every word themselves.

The evidence on whether AI detectors can be trusted to make that call is now substantial, and it is not reassuring.

The Accuracy Numbers

A 2026 study by researchers at Sultan Qaboos University found leading AI detection tools achieved overall accuracy of only 69 percent and 61 percent.

On hybrid text — writing where a person and an AI both contributed, which is now how a great deal of writing is actually produced — performance fell to nearly zero.

The vendor claims and the independent tests do not agree. Turnitin has stated its AI checker has a false-positive rate below 1 percent. A study by the Washington Post produced a much higher rate — 50 percent.

Who Gets Wrongly Accused

This is the finding that should concern every educator in a country where students write in English as a second language.

A study by Liang and colleagues found AI detectors misclassified more than 61 percent of essays written by non-native English speakers as AI-generated, while achieving near-perfect accuracy on essays by native speakers. Across seven detectors, the false-positive rate on TOEFL essays was 61.3 percent, against close to zero for native writing.

The likely mechanism is uncomfortable. Detectors look for predictable, lower-variety word choices. Careful second-language writers, using the vocabulary they are confident in, produce exactly that pattern.

For Bangladeshi students — whether studying at home or applying to universities abroad — that is not an abstract statistic.

Why Detection Is So Hard

Detectors do not find proof. They output a probability, and how reliable that probability is depends on the benchmark, the threshold chosen, the subject area, and whether the text has been revised.

A large-scale evaluation using the RAID benchmark showed detector performance changes substantially across different AI models, different writing domains and simple adversarial edits. Light rewording can defeat a detector; careful original writing can trigger one.

Institutions Are Changing Course

In 2026 a New York court reversed a student's expulsion that rested on a false Turnitin flag, and a number of colleges have disabled AI detection features.

Policy is moving in both directions at once: some institutions are tightening screening while others are removing it entirely.

If You Are a Student Who Was Flagged

1. Ask what evidence exists beyond the detector score. A probability is not proof.
2. Show your process. Drafts, version history in your document editor, notes and sources are strong evidence of authorship.
3. Offer to discuss the work. Explaining your own argument in conversation is among the most convincing demonstrations that you wrote it.
4. Cite the research. The findings above are published and can be shared with an academic panel.

If You Are a Teacher

Never treat a detector score as the sole basis for an accusation. The evidence shows it is least reliable for exactly the students most likely to be writing in a second language.

More durable approaches exist: assessing drafts and process rather than only the final text, in-class writing, oral discussion of submitted work, and assignments tied to local context and personal reflection that generic AI output handles poorly.

The deeper question for schools is not how to catch AI use. It is what students should be learning in a world where the tools exist — the same question Bangladesh's AI olympiad winners suggest the country is capable of answering well.

Related reading

• Why AI Makes Things Up, and a Five-Step Way to Check It
• Asking AI About Your Health? The Risks, and How to Do It Safely
• How Well Does AI Understand Bangla? A Practical Guide
• AI and Copyright: 130 Lawsuits and a $1.5 Billion Settlement

Sources

• "AI detector false positives: 2026 update," CASRAI — casrai.org
• "How do professors detect AI in 2026? Tools, accuracy, and false positives," Thesify — thesify.ai
• "The problems with AI detectors: false positives and false negatives," University of San Diego Legal Research Center — sandiego.edu
• "Why AI-generated text detection fails: evidence from explainable AI beyond benchmark accuracy," arXiv — arxiv.org

What Happens to What You Type Into an AI Chatbot

Free AI chatbot tiers generally use your conversations for training unless you switch it off, while business and enterprise plans exclude training by default. Once data enters a training run it cannot be removed.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

People paste astonishing things into AI chatbots: contracts, medical results, business plans, customer lists, passwords. Very few have checked what happens next.

The Default Rules in 2026

The pattern across the major providers is consistent enough to state simply.

Free tiers generally use your conversations to train future models unless you opt out.

Business and enterprise plans exclude your data from training by default. ChatGPT Team and Enterprise accounts, for example, are excluded from model training, and enterprise tiers typically come with compliance certifications such as SOC 2.

Personal paid plans vary by provider and have changed over the past few years. Do not assume — check the setting.

You Can Opt Out, Everywhere That Matters

ChatGPT, Claude, Gemini and Grok all provide settings to stop your conversations being used for training. The option usually sits under privacy or data controls in account settings.

Two limits are important:

• Providers may still keep logs for a limited period for safety review and legal reasons, even when training is off.
• Once information has been included in a training run, it cannot be removed retroactively. The earlier you opt out, the more of your data stays out.

The Protection That Does Not Exist

As of 2026, no major chatbot offers end-to-end encryption — the arrangement where only you hold the key and the provider cannot read your content.

That is the correct mental model to keep: anything you type into a cloud AI service is readable by that service. Settings control what they do with it, not whether they can see it.

Six Rules Worth Following

1. Turn off training today on every AI account you use.
2. Never paste passwords, API keys or banking details. There is no task that needs them.
3. Remove names and identifiers before pasting customer, patient or employee data. "Customer A" works just as well for most tasks.
4. Use a business plan for business data. If your company handles client information, the free tier is the wrong tool.
5. Check your organisation's policy before using AI at work. Using unapproved tools — "shadow AI" — is a named security risk.
6. For genuinely sensitive material, use a local model. A model running on your own computer never sends the data anywhere.

The Hidden Risk: Connected Tools

Privacy is no longer only about what you type. AI tools increasingly connect to your email, files and calendar through protocols such as MCP. Every connection gives the tool access to more of your data.

Review connected apps as carefully as you would review which apps can read your phone's contacts. Disconnect the ones you no longer use.

Why This Matters for Bangladeshi Users and Businesses

Bangladeshi freelancers, software exporters and outsourcing firms routinely handle confidential information belonging to foreign clients. Pasting a client's data into a free AI tool that trains on conversations can breach a contract — and damage a reputation built over years.

A clear AI data policy is now part of professional credibility. Clients are starting to ask for one.

Related reading

• Anthropic Has Overtaken OpenAI in Business AI Spending
• Asking AI About Your Health? The Risks, and How to Do It Safely
• How Much Electricity and Water Does AI Actually Use?
• What an AI Agent Actually Is, and How It Differs From a Chatbot

Sources

• "How to opt out of LLM training data: ChatGPT, Claude, Grok, Gemini and more (2026)," TrustScan — trustscan.dev
• "Are your AI chats private? How to opt out on every platform," LumiChats — lumichats.com
• "AI privacy: what really happens when you chat with AI," Prompt20 — prompt20.com
• "How to stop AI from training on your data: the 2026 privacy guide," Fello AI — felloai.com

AI Is Not Replacing Jobs Evenly. It Is Splitting Them in Two.

Freelancers doing complex work with AI saw earnings rise 45% in a year, while template design, commodity copywriting and data entry are shrinking. Which work AI is replacing, which it is raising, and how to tell yours apart.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

It is the most searched question about artificial intelligence, and it is usually answered with either reassurance or panic. The 2026 evidence supports neither. It shows something more specific: AI is dividing work into the part it replaces and the part it makes more valuable.

The Scale

Expert estimates suggest around 60 percent of occupations could be affected by AI by 2030. "Affected" is the key word — it means some tasks inside a job change, not that the job disappears.

Readers should treat long-range forecasts cautiously. Predictions about AI and employment have a poor track record in both directions. The more useful evidence is what is already happening.

What Is Actually Shrinking

Across creative and administrative work, the declining categories are consistent:

• Stock photography.
• Template-based graphic design.
• Commodity copywriting — product descriptions, generic blog content.
• Basic video editing.
• Data entry, with one industry projection estimating 65 to 80 percent of current roles gone by 2028.

In customer service, Klarna has reported its AI handling around 70 percent of customer interactions, with a reduction of roughly 700 agents in customer service headcount.

The common thread is standardised output. Work where the product looks broadly the same every time is the work AI does cheaply.

What Is Growing

Upwork's Future Workforce Index 2026, based on 2,400 skilled US knowledge workers plus platform data, found:

• More than 1 in 3 skilled knowledge workers now freelance, up from roughly 1 in 4 a year earlier.
• Freelancers doing more complex work with AI saw earnings rise 45 percent year on year.
• AI-augmented professional services — domain experts using AI inside established fields — grew 72 percent in volume, with earnings rising.

Growing roles include AI creative directors, AI-augmented senior designers and content strategists. Salary trends across creative work show what analysts call extreme bifurcation: the top rising, the bottom falling.

How to Tell Which Side Your Work Is On

Ask four questions about the tasks you spend most of your week on.

1. Is the output standardised? If a good result looks roughly the same each time, it is exposed.

2. Does it require judgement about a specific situation? Knowing this client, this market, this legal context is harder to automate.

3. Does someone need to trust you? Accountability, relationships and responsibility for outcomes remain human.

4. Can you check whether the AI got it right? People who can evaluate AI output are needed precisely because, as our guide to AI errors sets out, it is often confidently wrong.

What It Means in Bangladesh

Bangladesh has one of the world's largest online freelance workforces, and a significant share of it has competed on exactly the standardised tasks now shrinking: basic design, data entry, simple content and routine development.

That is a real risk, and it should be said plainly. It is also a clear signal about where to move. The same data shows complex, AI-augmented work paying more, not less. Our practical guide for freelancers covers how to make that shift.

For a country where 65 percent of people are under 35, how quickly the workforce moves from standardised output to judgement-based work is one of the most important economic questions of the decade.

The Honest Answer

AI is unlikely to take most jobs outright in the near term. It is very likely to take the most repetitive parts of many jobs, and to reduce what people are paid for work that stays standardised.

The people doing best are not avoiding AI. They are the ones using it to do the complex part of their job faster.

Related reading

• AI for Freelancers: The Work That Still Pays, and How to Move Into It
• How Much Electricity and Water Does AI Actually Use?
• GPT-6 Astra and What It Means for Bangladesh
• Is Your Data Safe With AI Chatbots? How to Opt Out of Training

Sources

• "Upwork's Future Workforce Index 2026: how AI is redefining the value of work as skilled freelancing accelerates," Upwork — upwork.com
• "The 2026 AI job disruption report: which roles are being eliminated, which are being created," AI Magicx — aimagicx.com
• "60+ AI job replacing statistics relevant for 2026," Tenet — wearetenet.com
• "Labor market impacts of AI: a new measure and early evidence," Anthropic — anthropic.com

From Noise to Picture: How AI Draws, and How to Direct It

Diffusion models build an image by shaping random noise step by step. Here is how Midjourney, FLUX, Stable Diffusion and GPT Image compare in 2026, and a prompt structure that gets usable results.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 7-minute read

Type a sentence, get a picture. Behind that simple exchange is a process that is genuinely strange, and understanding it makes you much better at getting the image you had in mind.

How Diffusion Works

Most leading image generators use diffusion. The model starts with a canvas of pure random noise — television static — and removes noise in many small steps, each time nudging the pattern toward something that matches your description.

It learned to do this by studying the reverse: taking real images, adding noise until nothing recognisable remained, and learning how to undo each step.

Doing this on full-resolution pixels would be enormously expensive. Systems such as Stable Diffusion work in a latent space instead — a compact, simplified map of possible images — and only convert to full pixels at the end.

Not Only Diffusion Any More

In 2026, open image generation includes diffusion, autoregressive and hybrid architectures, each tuned for different priorities: accurate text inside images, generation speed, editing workflows, or knowledge of real-world subjects.

Accurate lettering was for years the obvious weakness — AI signs full of nonsense characters. It is now a specific target of newer models.

The Main Tools Compared

Midjourney V8.1 is widely regarded as the leader for artistic style.

GPT Image 2 and Google's Nano Banana models sit at the top of current head-to-head quality rankings.

FLUX.2 scores around 1,190 Elo in those rankings — behind GPT Image 2 and Nano Banana, ahead of Imagen 4 and well ahead of Stable Diffusion 3.5. It is favoured for realistic images by photographers, content creators and marketers.

Stable Diffusion, from Stability AI, remains the major open-source family, with versions from 1.4 through XL to 3.5 Large. Its advantage is that you can run and modify it yourself.

Platforms such as NightCafe give access to several models in one place, with a community and credit system.

A Prompt Structure That Works

Vague prompts get generic pictures. Build the prompt from six parts:

1. Subject. Exactly what is in the picture. "A tea picker in a Sylhet tea garden" beats "a worker".
2. Setting. Where and when. Morning mist, monsoon light, a crowded Dhaka street.
3. Style. Photograph, watercolour, flat illustration, editorial photography.
4. Lighting. Soft window light, golden hour, overcast. Lighting changes a picture more than almost anything else.
5. Composition. Close-up, wide shot, from above, subject on the left third.
6. Format. Aspect ratio — 16:9 for a website banner, 1:1 for social media, 9:16 for a phone screen.

Then iterate. Change one element at a time, so you learn which word caused which change.

Common Problems and Fixes

• Hands and fingers look wrong: reframe so hands are less prominent, or regenerate.
• Text is garbled: use a model known for typography, or add the text afterwards in an editor.
• Everyone looks the same: describe age, clothing and setting specifically — models default to the most common patterns in their training data.
• The style drifts across a series: reuse the exact same style phrase in every prompt.

The Rules Before You Publish

Label it. Readers deserve to know an image is AI-generated, especially near news. This publication marks AI images plainly, as in the caption above, and does not use them to depict real events.

Do not imitate real people or living artists. The legal position on style imitation and training data is still being fought in court — see our guide to AI copyright.

Check commercial terms. Each service sets its own rules for commercial use, and those rules differ.

Related reading

• How to Write an AI Prompt in 2026: What Actually Changed
• How Well Does AI Understand Bangla? A Practical Guide
• RAG Explained: How to Make AI Answer From Your Own Documents
• MCP Explained: The Protocol That Connects AI to Your Tools

Sources

• "The best open-source image generation models in 2026," BentoML — bentoml.com
• "The 9 best AI image generation models in 2026," Gradually.ai — gradually.ai
• "Diffusion models and image generation: from noise to reality," Weskill — blog.weskill.org
• "40+ best AI image generation tools for 2026," Dokan — dokan.co

A Million-Token Memory That Still Forgets Page 600

Thirteen AI models now advertise context windows of a million tokens or more, some up to 10 million. On every model benchmarked, the length where quality actually holds is shorter than the advertised figure.

By the UISC BD Editorial Desk · United Information Service Center · Published 13 September 2026 · 6-minute read

You start a conversation with an AI tool and it is sharp. Two hours and forty messages later it contradicts itself, forgets an instruction you gave at the start, and seems to have lost the thread.

It is not tired. It has run into the limits of its context window.

What a Context Window Is

A model has no memory between requests. Everything it knows about your conversation is the text sent to it each time: your instructions, the documents you shared, and the whole chat so far.

The context window is the maximum amount of that text, measured in tokens, a model can take in at once. Once a conversation exceeds it, something has to be dropped or summarised.

How Big They Have Become

In 2026, context windows range from about 128,000 tokens on standard models to 10 million on models such as Llama 4 Scout and Gemini 3 Pro. Thirteen models now ship windows of 1 million tokens or more — enough to hold several long books at once.

That sounds like the problem is solved. It is not.

Advertised Versus Effective

The effective context — the length at which quality actually holds — falls short of the advertised maximum on every model ever benchmarked.

The illustration used by one 2026 comparison is worth quoting in substance: load a 900,000-token document into a million-token model and ask about page 600, and the model may confidently give a wrong answer. The information is present. Its attention has drifted.

Lost in the Middle

The failure is not random. Models recall material at the beginning and end of a long context more reliably than material in the middle.

That has direct practical consequences:

• Put the most important instruction at the start, and repeat it near the end of a long prompt.
• Put the question after the document, not before it.
• Do not bury a critical fact in the middle of a 200-page paste and assume it will be found.

Why Bigger Is Not Always Better

Cost. Filling a 1-million-token window costs about $0.14 on DeepSeek V4 Flash and about $10.00 on Claude Fable 5 — a 71-times spread. Architectural techniques such as sparse attention, ring attention and efficient KV-cache management make huge windows possible, but they do not make them free.

Speed. More input takes longer to process before the first word of an answer appears.

Accuracy. As above, more is not the same as better remembered.

What to Do Instead

Start new conversations more often. When a thread drifts, open a fresh one with a short summary of what matters. This is the single most effective habit.

Send the relevant part, not everything. Retrieval — pulling only the passages that matter — usually beats stuffing a whole archive into the window. That is the approach behind RAG.

Keep standing instructions short. A three-page system prompt spends your context and your budget on every single message.

Verify answers from long documents. Treat them with the care described in our checking routine.

The Mental Model to Keep

A context window is a desk, not a filing cabinet. A bigger desk holds more paper, but the more you pile on, the harder it is to find the page you need. The skill is keeping the right papers on the desk.

Related reading

• What a Token Is, and Why AI Prices Fell About 80% in a Year
• MCP Explained: The Protocol That Connects AI to Your Tools
• How to Run AI on Your Own Computer: Ollama and llama.cpp
• How Well Does AI Understand Bangla? A Practical Guide

Sources

• "AI model context window comparison 2026: advertised vs. real," elvex — elvex.com
• "LLM context window 2026: 128K to 10M tokens — which to use," TokenMix — tokenmix.ai
• "LLM context windows explained: 4K to 1M tokens (2026)," DevTk.AI — devtk.ai
• "LLMs with largest context windows," Codingscape — codingscape.com