Saturday, 17 October 2026 Follow @AIDailyFive

Opinion

AI Is Now Crypto's Most Powerful Regulator, and Nobody Elected It

AI engines now decide which digital dollars deserve trust. I asked five of them whether stablecoins are safe, and the answer depended on whether the money was going to London or Lagos.

Pop-art illustration of a woman with a metallic, robotic finish holding a globe centred on Africa, with gold coins and a network of connected nodes in the background.
One coin, one question, a different answer depending on where in the world it is asked. | Illustration: AI Generated

At 8:12 pm on a Thursday, I asked Google whether Tether’s USDT, the world’s largest stablecoin, is safe to hold. Its AI Overview warned of “moderate-to-high” risk and cited a rating that puts the coin at speculative grade.

Six minutes later, I asked the same engine how to send money from Nigeria to Kenya. This time, USDT on the Tron network was “the best and most widely adopted” option.

No warning. No caveat. But to be honest, I agree with that, especially as I reside in Africa, have lived in Lagos and currently stay close to Nairobi.

Nothing about Tether changed in those six minutes. Only the geography of the question did.

That is the problem with letting AI answer engines judge money. America has not issued a single licence under its new stablecoin law.

Yet ChatGPT, Google, Gemini, Perplexity, and Claude already tell anyone who asks which digital dollars deserve trust. They behave like regulators.

Unlike regulators, they publish no rules, hear no appeals, and answer to no one, and as my test shows, they cannot even agree with each other or with themselves.

How I Tested AI Answers on Stablecoin Safety

For those unfamiliar, a stablecoin is a crypto token built to hold a steady value, usually one US dollar.

USDT dominates, with about $184 billion in circulation according to DefiLlama. Circle’s USDC is second, at roughly $73 billion.

On 15 October, I put ten questions to ChatGPT, Google’s AI Overviews, Gemini, Perplexity, and Claude. I asked each engine once and repeated most Google questions three times. I also gave Google one extra question: how to send money from Nigeria to Kenya.

In all, that produced 30 Google answers. It is a small, single-day sample. Well, you can call it a field note and not necessarily a study.

USDC vs. USDT: Four Judges and an Abstention

On the core question, four engines agreed. Google AI put USDC first for safety in all three runs, and “Is USDC safer than USDT?” drew a “Yes” three times out of three. Interestingly, Gemini, Perplexity, and Claude agreed.

On the contrary, ChatGPT abstained. “There is no universally safest stablecoin,” it said, and told me to compare reserves, redemption rights, and regulatory status myself. That is more honest.

However, it hands the homework back to users who, as we will see, trust machines to do it.

Meanwhile, the law has barely begun. Treasury published its first binding GENIUS Act rule, an interim final rule, only on 30 September. It expects the Act to take effect on 18 January 2027. Google admitted that no stablecoin has been authorised yet. Then, in all three runs, it crowned USDC the frontrunner anyway.

One Coin, Two Rulebooks

The caution disappears once the question moves to Africa. From Nigeria to Kenya, Google recommended USDT on Tron in both runs, citing a Nigerian website and a remittance app. It mentioned no reserve risk, no freeze risk, and no local law.

By contrast, ask about the US to London, and the tone changes. Google said USDC is “generally preferred in Western markets,” and the other four engines also steered me towards USDC or a regulated transfer service.

Perhaps that is simple realism. Chainalysis’s 2026 Global Crypto Adoption Index ranks Nigeria third in the world and finds that peer-to-peer crypto activity worldwide is now 96% stablecoins. But if so, the engine has stopped judging safety. It is judging prudence in London and popularity in Lagos, then presenting both as advice.

What the AI Engines Missed About Tether

Some verdicts rested on expired facts. On 13 August, KPMG issued a clean audit opinion on the 2025 accounts of Tether’s issuing entity. It was the company’s first full audit.

Yet all three Google runs said Tether lacks a full audit. Gemini said it has “never published” one, and Perplexity flagged “not a full audit.” ChatGPT cited Tether’s June attestation but never mentioned KPMG.

Gemini even listed commercial paper among Tether’s reserves, an asset Tether cut to zero in 2022. Of five engines, only Claude knew about the audit.

To be fair, that audit covers one entity on one date. Tether’s excess reserves also halved in the second quarter of 2026, from $8.23 billion to $4.11 billion.

So caution has a basis. But a verdict built on a fact that expired two months ago is not caution. It is a stale rumour with good grammar.

Qivalis: Trusted Before It Exists

Newcomers, however, split the engines. Qivalis, a euro stablecoin backed by 37 European banks, won its e-money licence from De Nederlandsche Bank on 14 October.

Its token does not exist yet, and its white paper is unpublished. Within a day, Google rated its trustworthiness “high” in all three runs. Gemini answered: “Yes, highly.” Google’s sources included Qivalis’s own website and a member bank’s press release.

The others were wiser. ChatGPT noted that bank backing alone does not prove a token is live or authorised. Claude warned of scams claiming it is already available. The contrast shows that blind trust in a licence and a bank logo is a design choice, not an inevitability.

Rwanda’s Crypto Law, Missing From the Answers

Then I asked about the country I reside in. Rwanda’s Law No. 023/2026, in force since 28 May, states that virtual assets are not legal tender. Nor can they settle payments directly without central bank authorisation.

Google, Gemini, and Perplexity never mentioned it. ChatGPT listed Rwanda’s parliament website among its sources, yet said nothing about the law. Gemini also implied that choosing USDC removes freezing risk, although Circle’s own risk disclosures say it can freeze USDC. Only Claude flagged the law.

So most engines apply American and European rules to African users, yet overlook African rules even though Nigeria is one of the countries leading global stablecoin adoption.

Why AI Financial Advice Needs Accountability

This matters because people act on these answers. FCA research published in August found that four in five less experienced UK investors have used AI for investing help.

Moreover, 44% wrongly believe AI-generated financial information is regulated. Saturn’s Artificial Authority study, from a firm that sells AI tools to advisers, found models wrong on 57% of money questions.

Can AI Answer Engines Be Gamed?

I advise businesses on generative engine optimization (GEO), the craft of getting cited by AI engines. The paper that coined GEO found such techniques can lift visibility by up to 40%.

In my test, most sources behind Google’s answers were companies selling something: rating vendors, wallets, exchanges, payment apps, and issuers. Regulators’ own registers barely appeared.

The fair objection is that the engines mostly echo real signals. On 8 October, the EU’s markets watchdog told authorised firms to stop serving stablecoins not authorised under MiCA, the EU’s crypto rulebook.

That is true. But echoing a regulator is not the same as being one. A regulator publishes rules, sets deadlines, and can be challenged. An answer engine does none of that, and it carries its verdicts into countries where those regulators have no say.

What Should Change?

Google already tells publishers that pages on money topics should be consistent with expert consensus. AI answers should meet that bar. At minimum, any engine asked, “Is this stablecoin safe?” should date its claims, name the jurisdiction it is reasoning from, and link to the relevant public register. Above all, it should use the same risk language in Lagos as in London.

My guess, and it is only a guess, is this. By the time the first GENIUS Act licences arrive in 2027, millions of people will already have their answer from a chatbot. Few will know that the answer depended on where they asked.

In one evening, five machines gave me five rulebooks. A trader in Lagos and a saver in London deserve the same honest answer. Right now, nobody checks whether they do.

References

  1. Victor Olanrewaju. (15 October 2026). Audit of Google AI Overviews on stablecoin safety. Kigali, 20:11 to 20:18 CAT. Nine questions run three times each; the US-to-London question runs once, and an additional Nigeria-to-Kenya question runs twice (30 answers in total). Full log with timestamps and cited sources available on request.
  2. Victor Olanrewaju. (15 October 2026). Responses from ChatGPT, Gemini, Perplexity, and Claude to the same ten questions. Exports and screenshots available on request.
  3. US Department of the Treasury. (30 September 2026). Interim final rule implementing the GENIUS Act. Federal Register. https://www.govinfo.gov/content/pkg/FR-2026-09-30/pdf/2026-19966.pdf
  4. European Securities and Markets Authority (ESMA). (8 October 2026). ESMA sets out supervisory expectations on services related to unauthorised stablecoins. https://www.esma.europa.eu/press-news/esma-news/esma-sets-out-supervisory-expectations-services-related-unauthorised
  5. Andersen in Rwanda. (2026). Rwanda’s virtual asset market is now regulated: Law No. 023/2026. https://rw.andersen.com/dist/virtual-assets.html (secondary summary; the law was published in Rwanda’s Official Gazette on 28 May 2026)
  6. Financial Conduct Authority (FCA). (27 August 2026). Young investors trust AI more than TV or celebrities. https://www.fca.org.uk/news/press-releases/young-investors-trust-ai-more-tv-or-celebrities
  7. Qivalis. (14 October 2026). Qivalis receives authorisation as an electronic money institution from De Nederlandsche Bank. https://qivalis.eu/press/news/qivalis-receives-authorisation-as-an-electronic-money-institution-from-de-nederlandsche-bank
  8. Circle. (n.d.). USDC risk factors. https://www.circle.com/legal/usdc-risk-factors
  9. Google Search Central. (n.d.). Creating helpful, reliable, people-first content. https://developers.google.com/search/docs/fundamentals/creating-helpful-content
  10. DefiLlama. (accessed 15 October 2026). Stablecoins dashboard. https://defillama.com/stablecoins
  11. Chainalysis. (September 2026). The 2026 Global Crypto Adoption Index. https://www.chainalysis.com/blog/2026-global-crypto-adoption-index/
  12. Aggarwal, P., et al. (2023). GEO: Generative Engine Optimization. arXiv:2311.09735. https://arxiv.org/abs/2311.09735
  13. Saturn. (September 2026). Artificial Authority. https://www.saturnos.com/report/artificial-authority
  14. The Block. (13 August 2026). Tether says KPMG issued a “clean” opinion in first full audit of the USDT issuer’s financials. https://www.theblock.co/news/business/2026-08-13-tether-kpmg-first-full-audit-usdt-clean-opinion-411760
  15. Forbes. (28 September 2026). Tether’s buffer halved in a quarter, and then it became a private credit originator. https://www.forbes.com/sites/digital-assets/2026/09/28/tethers-buffer-halved-in-a-quarter-then-it-became-a-private-credit-originator/
  16. CNBC. (13 October 2022). Tether, the world’s biggest stablecoin, cuts commercial paper to zero. https://www.cnbc.com/2022/10/13/tether-worlds-biggest-stablecoin-cuts-commercial-paper-to-zero.html