The first independent safety evaluation of OpenAI's ChatGPT Health feature, published in Nature Medicine, found the tool failed to direct users to emergency care in 51.6% of cases requiring immediate hospitalization - instead recommending they stay home or book a routine appointment. The study also found ChatGPT Health frequently failed to detect suicidal ideation, with suicide crisis alerts sometimes triggering in lower-risk scenarios while failing to appear when users described specific plans for self-harm. Over 40 million people reportedly ask ChatGPT for health-related advice every day.

The U.S. Court of Appeals for the Fifth Circuit sanctioned attorney Heather Hersh $2,500 after finding her brief contained 16 fabricated quotations and five additional serious misrepresentations of law or fact, all apparently AI-generated. The court expressed frustration that AI-hallucinated legal citations "have increasingly become an even greater problem in our courts" and that the issue "shows no sign of abating." Hersh initially denied using AI, then shifted to claiming she "relied on publicly available versions of the cases, which she believed were accurate."

Maryland attorney Kusmin Amarsingh used ChatGPT to draft her appellate brief against Frontier Airlines without verifying any citations, resulting in multiple nonexistent cases being cited in the 10th Circuit. The court found her conduct "reckless" for completely failing to perform "an attorney's fundamental duty to the court." She was fined $1,000 and referred to Maryland attorney-disciplinary authorities.

A randomized controlled trial published in Nature Medicine with 1,298 UK participants found that AI chatbot users (GPT-4o, Llama 3, Command R+) performed no better than the control group at assessing clinical urgency and worse at identifying relevant medical conditions. In one case, two users with identical subarachnoid hemorrhage symptoms received opposite recommendations -- one told to lie down in a dark room, the other correctly advised to seek emergency care.

Attorney Steven Feldman filed multiple motions containing AI-fabricated case citations in Flycatcher Corp. v. Affable Avenue LLC. Despite explicit court warnings and access to Westlaw and Lexis, he continued submitting unverified AI output -- even using AI to draft his response to the court's show-cause order, which contained yet more fake citations. Judge Failla imposed the most severe AI-hallucination sanction yet: default judgment against his client.

A federal judge in the District of Kansas fined four attorneys a combined $12,000 for court filings containing AI-generated fabricated legal citations in a patent infringement case. The attorney who used ChatGPT received $5,000; two who failed to review the filings received $3,000 each; local counsel who did not identify errors received $1,000. The judge called the volume of fabricated case law "staggering."

Attorneys Yen-Yi Anderson and Jeffrey Goldin jointly filed a motion in Lifetime Well v. IBSpot containing at least eight AI-generated false citations. Judge Kearney imposed differential sanctions based on their responses: Anderson, who blamed time pressure and fired her law clerk rather than accepting responsibility, received $4,000 in monetary sanctions. Goldin, who promptly accepted responsibility and implemented remedial measures, received no monetary penalty.

A New York appellate court imposed $10,000 in sanctions after a lawyer submitted briefings in a mortgage foreclosure case containing fabricated case citations identified as likely AI-generated hallucinations. The court found multiple nonexistent cases and misrepresented holdings, affirming prior orders and awarding costs to the plaintiff.

Five attorneys who signed a legal brief in McPhaul v. College Hills submitted fabricated case citations hallucinated by ChatGPT to a federal court in Kansas. The judge issued an order requiring them to explain why they should not be sanctioned, with multiple defects attributed to AI in the documents.

Acquire BPO’s 2024 AI in Customer Service survey found 70% of U.S. consumers would bolt to a rival after just one bad chatbot interaction and 72% only buy when a live agent safety net exists, even as CMSWire reports enterprises poured $47 billion into AI projects in early 2025 that delivered almost no return. CX strategists now warn executives that Air Canada–style hallucinations, mounting legal liability, and empathy gaps make AI-only helpdesks a churn machine unless human agents stay in the loop.

A BBC audit of 2,700 news questions asked in 14 languages found that Gemini, Copilot, ChatGPT, and Perplexity mangled 45% of the answers, usually by hallucinating facts or stripping out attribution. The consortium logged serious sourcing lapses in a third of responses, including 72% of Gemini replies, plus outdated or fabricated claims about public-policy news, reinforcing fears that AI assistants are siphoning audiences while distorting the journalism they quote.

Conservative organizer Robby Starbuck sued Google in Delaware, saying Gemini and Gemma kept spitting out fabricated claims that he was a child rapist, a shooter, and a Jan. 6 rioter even after two years of complaints and cease-and- desist letters. The $15 million suit argues Google knew its AI results were hallucinated, cited fake sources anyway, and let the libel spread to millions of voters.

Deloitte admitted AI-generated errors in a commissioned Australian government report and agreed to refund the fee.

Los Angeles attorney Amir Mostafavi became the first California lawyer sanctioned for AI-generated legal fabrications when a court hit him with a $10,000 fine. He ran his appeal draft through ChatGPT to improve the writing but did not verify the output before filing, unaware the tool had inserted fabricated case citations.

A Washington patient replaced table salt with sodium-bromide after ChatGPT said it was a healthier substitute. The patient developed bromism and psychosis, resulting in a hospital stay that doctors now cite as a warning about AI health guidance.

After the Air India 171 crash, synthetic images and AI-generated claims spread widely, confusing even experts.

A King Features syndicated summer reading list used AI and included nonexistent books. It appeared in the Chicago Sun-Times and one edition of the Philadelphia Inquirer before corrections and apologies.

Meta expanded its AI assistant across apps, then limited it after high-profile bad answers - including on breaking news.

Google’s AI search overviews went viral for bogus answers, including telling people to add glue to pizza and eat rocks.

NYC’s Microsoft-powered MyCity chatbot gave inaccurate/illegal advice on labor & housing policy; city kept it online.

Attackers register software packages that AI tools hallucinate (e.g. a fake 'huggingface-cli'), turning model guesswork into a new supply-chain risk dubbed "Slop Squatting".

Google paused Gemini’s image generation of people after it produced inaccurate historical depictions and odd refusals.

Tribunal ruled Air Canada responsible after its AI chatbot misled a traveler about bereavement refunds.

Gannett halted Lede AI high-school recaps after robotic, error-prone stories went viral.

In Mata v. Avianca, attorneys submitted a brief citing non-existent cases generated by ChatGPT. A federal judge sanctioned two lawyers, ordered a $5,000 penalty, and required notices to judges named in the fake citations.

In its launch promo, Bard claimed JWST took the first exoplanet photo - which was false. The flub overshadowed the event and dented confidence.

CNET paused and reviewed AI-generated money articles after multiple factual errors were found.