Insights
Research on how AI describes people and companies.
2026-10-10
A US court ordered the Army's AI evaluations of TRAX and two rival bidders into the record, though the government said they were never in the contract file. TRAX lost its protest.
2026-10-09
The ICO secured data changes from OpenAI, Google, Meta and seven other AI developers on 8 October 2026. Its expectations say nothing about accuracy or correcting false outputs.
2026-10-07
On 30 September 2026 an Arizona appeals court vacated the Horcasitas sentence, ruling an AI video of victim Christopher Pelkey unreliable because it documented nothing real.
2026-10-06
A Munich court barred Google from repeating AI Overview claims linking two publishers to scams, holding the summaries are Google's own statements. Google plans to appeal.
2026-10-01
Three benchmark studies this month found AI shopping assistants inventing product specs and prices, then contradicting themselves minutes later, with no record of what was said or to whom.
2026-09-25
A federal judge let a defamation claim against Google's AI Overviews proceed -- not because the AI was wrong, but because someone could prove Google was told and kept publishing anyway.
2026-09-22
Insurers are excluding AI risk from D&O and liability policies. Coverage is coming back only for companies that can document what an AI system said, and when.
2026-09-19
Courts now treat AI chat logs as ordinary discoverable evidence. There is still no comparable way to prove what an AI system said about someone else, on a given day, before it changed.
2026-09-17
Anthropic named the targets in a September disclosure. The named organisations had no say in the story AI systems now tell about it.
2026-09-13
China's top court just wrote the first liability rules for AI hallucinations — but left open exactly what counts as proof of what a system said, and when.
2026-09-10
AI-related securities suits were 13% of H1 2026 filings but 73% of alleged losses — and the fights are increasingly about timing, not truth.
2026-09-04
AI systems debunk foreign propaganda in English about 75% of the time. Ask about the same governments in their own language, and the answers turn measurably friendlier.
2026-09-01
Nearly 2,000 court rulings now document AI-hallucinated legal filings, and each one becomes a permanent record AI systems can surface about the lawyer or firm involved.
2026-08-31
Up to 93% of AI-mode searches now end without a click to a source. Due diligence increasingly stops at the chat window, with no record of what was said or where it came from.
2026-08-28
Nearly 3,750 AI-run content mills are already funded by advertisers who do not know it, and increasingly commissioned to fabricate damaging claims about named companies.
2026-08-25
A Minnesota company is suing Google after its AI Overview invented a regulatory lawsuit that never existed, built from four real sources that said nothing of the kind.
2026-08-22
Researchers tested eight AI search engines on the same task and found error rates from 37% to 94% -- delivered with almost no hedging. Ask one where a claim about you came from, and the answer may be invented.
2026-08-21
A fabricated citation, once published, becomes indistinguishable from a real one the moment it is cited again. The same laundering mechanic is now turning invented claims about people and organisations into apparent fact.
2026-08-18
A German court just ruled a company can't blame its own chatbot for lying to customers. The precedent has implications far beyond chatbots companies actually operate.
2026-08-15
Half of dealmakers now run AI systems before the first call. The profile it produces is stale, uncited, and rarely saved -- so no one can prove what shaped the decision.
2026-08-05
New EU transparency rules force AI systems to disclose they're AI. They say nothing about whether what those systems claim about you is true.
2026-07-30
A Delaware court let a defamation claim over Gemini's outputs proceed, on the strength of a reach figure the chatbot supplied about itself.
2026-07-27
AI systems still rank a 1982 recall as the top crisis case study. New research shows why some crisis narratives compound for decades — and why the record matters more than ever.
2026-07-24
The first wave of AI defamation suits isn't failing because judges believe AI systems get it right. It's failing because almost nobody can prove who saw the false claim, or what it cost them.
2026-06-18
When a public AI system researches your company, it reads pages it doesn't own — and a hidden line of text can quietly rewrite what it reports about you.
2026-06-15
Banks, acquirers and partners now run automated screens powered by public AI systems. When those systems invent adverse news, a company can fail a test it never knew it took.
2026-06-09
More of what public AI systems read is now written by other AI. When a false claim about your organisation gets copied back in, repetition starts to look like fact.
2026-06-06
Ask five public AI systems whether a company is sound and you'll get five different answers. The gap between them is invisible to most leaders — and it's exactly where the risk hides.
2026-06-03
A state regulator caught a public AI system inventing a medical licence number. The detail is not a bug — it is how fabrication works, and the new rules will not fix it for you.
2026-05-21
Public AI systems now cite their sources. The problem? Those sources often don't support what the AI actually said about your organisation.
2026-05-18
Wikipedia is the single most cited source in ChatGPT responses. For most organisations, that page hasn't been checked in years.
2026-05-09
Every press release, earnings call, and investor deck you've published is potential source material for public AI systems. The context expires. The claim doesn't.
2026-05-06
Public AI systems describe organisations differently in different languages. Stakeholders in each market see a different version of the same company — and most leadership teams have no visibility into the gap.
2026-05-04
Regulatory staff are using AI research tools before formal examinations begin. Before the first letter arrives, a public AI system may already have shaped how examiners frame their questions about your organisation.
2026-04-30
The FTC brought a dozen AI washing cases in 2025. The SEC is actively scanning how companies describe their AI capabilities. Your past claims may still be circulating in public AI systems — without context.
2026-04-24
82% of journalists now use AI tools for research. Before your press team answers a single question, a reporter may already have a picture of your company built by a public AI system.
2026-03-30
Public AI systems are generating false claims about real companies — and the AI providers say it's not their problem. Who fills the gap?