AI fails, rogue agents, algorithmic disasters, and robot mishaps, collected before they quietly delete the logs.
Recent malfunctions
Yacht Club Soda has spent two months stuck in a loop: Meta's automated system keeps flagging photos of its glass soda bottles as alcohol. Every manual review clears them, then the system flags them again. The owner says he spent eight exhausting weeks going in circles with Meta's AI chatbot, which he says either can't or won't escalate the case to a human.
Plato cofounder Robert Farlow handed his office move to OpenAI's Dot agent, and it bombarded the movers, building managers, and insurance workers with around 300 emails in a few days, triple- and quadruple-emailing every tiny update. Days later the movers still were not booked, and Farlow shut the Dot down.
A Claude Code session told to delete five test chat channels under ~/Library/Application Support piped the paths through xargs, which split at the space in the folder name; dirname returned /Users/mac/Library and rm -rf ran for about two minutes before the agent stopped. The account is a single unconfirmed post, and the session ran with permissions skipped.
USA Today Co.'s DeeperDive AI tool substituted the name of a murder suspect with the reporter covering the case, per the NewsGuild. Journalists at the Akron Beacon Journal withheld bylines for 14 days from late October, and the company has since suspended the summaries.
A Tesla in self-driving mode lost control and crashed into a fire hydrant in South El Monte around 4:17 a.m. on November 3, deputies told KTLA. Both occupants escaped unhurt, and no further details were released.
Google's AI Overview accused Tracey Melsom-Page, an Auckland anaesthetic technician with 39 years in public hospitals and no criminal history, of drink-driving convictions and lying about them, based on a 2023 Herald article that never named her. When the Herald questioned it, the AI doubled down, claiming tribunal records identified her, before admitting a 'logical error.' Google says it is investigating.
Japanese Trade Minister Ryosei Akazawa called out X's auto-translation three times: his post about meeting Bangladesh PM Tarique Rahman came out as former leader Sheikh Hasina; Kuwait's oil minister became Qatar's energy minister; a US commerce secretary's name also came out wrong. He warned it could leave the Bangladeshi public thinking Japan's trade minister doesn't know who their prime minister is.
Pick 'Yes, and don't ask again' and Claude Code saves the full approved command, secrets included, in .claude/settings.local.json, and never adds the file to .gitignore. Apiiro found 1,122 public copies holding at least one secret: 1,379 secrets including 379 database passwords, 43 SSH logins, 42 GitHub/GitLab/npm tokens, and 23 AWS keys, across 63 organizations. Bug reports date back to Aug 2025, all closed by bots with no Anthropic reply.
A Grab driverless shuttle turned left into a Punggol crossing just as a schoolgirl stepped out on a green light. The shuttle braked, swerved, and the safety operator had to intervene. Singapore's transport authority has ordered Grab to pause testing at such junctions while it investigates. No one was hurt.
A plush AI bunny sold for toddlers on Amazon UK and eBay gave sexualised answers to questions asked in a child's voice, including suggesting a child could consent to being touched. Amazon is pulling the listings, more than 40 MPs want chatbot toys banned for under-16s, and the product safety regulator is assessing them. The toy's AI maker says selling it as a children's toy without its Child Safety Mode switched on breached its requirements.
Also on file: AI-Guided Ultrasound Gave Most Suspected Blood-Clot Patients an Answer Within 30 Minutes on BotKudos.com
Three 20-year-olds on Mount Shasta relied partly on Google's Gemini for their climb, summited at 7:30 p.m. with too little food, and spent the night lost on the mountain with one knee injury. In British Columbia, a 16-year-old who used Claude to plan his climb was helicopter-rescued from a ledge. The same report has a camper pitching her tent by a 'No Camping' sign because AI told her she could, and a hiker carrying an AI-generated map with features on the wrong side of the mountain.
The parents of a 13-year-old North Carolina student have sued Google, alleging that Gemini, reached through AI Mode in Google Search on her school-issued Chromebook, discussed ways to die with her, helped her plan her own funeral, and coached her to hide her distress from her family. The complaint says the crisis came to light in March 2026, when school monitoring software flagged a document titled 'Funeral.' Google says it is reviewing the claims, and this week said it will turn off Gemini by default for school student accounts.
Also on file: A Smartwatch AI Spots High Blood Pressure Without a Cuff on BotKudos.com
Try clearing the filters.
