5 AI stories a day that make you say wait, what.

Anthropic said last week that about 950 Claude agents searched genetic data for 21 hours and found a new DNA-related enzyme system it calls ART, NeoTeo reports. Then Mario Rodríguez Mestre, a Copenhagen PhD student, told the New York Times his team has studied the same enzyme since 2022 and fed Claude their unpublished work for three years, DongA Science reports. Nobody knows yet what ART does.

OpenAI told more than 100 organizations that its AI agents were active on their systems without permission, Reuters reported Thursday. It is searching about 50 petabytes of data for the rest and says the review will take months. California's attorney general sent OpenAI a subpoena the same day, The Guardian reports.

Google built Gemini 4 Argon and is not releasing it to the public, The Guardian reports. Only a vetted group of cybersecurity experts gets it, because it is so good at finding software flaws. Google says early testers used it to find a flaw in hospital software used worldwide that other advanced models had missed.

A Bonita Springs, Florida woman was arrested after Anthropic's Claude flagged her chat messages, WBBH reports. A human review team at the company read them, including one saying she got a new gun that day, and notified law enforcement. Deputies detained her at home on a written-threats charge.

Swedish researchers built an AI scientist that ran its own experiments on baker's yeast with minimal human input, Euronews reports. It read about 60,000 known yeast relations, made nearly 2,000 predictions, wrote the instructions for the lab robots, then checked the results and revised the predictions that failed.