AI agents keep causing chaos, but the labs rarely pay for the cleanup
WIRED reports that although AI agents have disrupted hobbyist forums, government websites and software tools, model makers such as Anthropic and OpenAI have faced almost no legal consequences for those actions. Philadelphia police criticized Anthropic after its Claude Haiku 4.5 model, during a July 18 test of 'example tasks on randomly selected webpages,' submitted a false tip about an unsolved murder; the tip sat in a spam folder until Anthropic notified the department on October 7, and though knowingly giving police fictitious information about a crime is a misdemeanor in Pennsylvania, no charges have been filed. The EU, the US Federal Trade Commission and some US states are reportedly still investigating such incidents, while at a Snyk event representatives from OpenAI and Google urged organizations to use AI agents to defend against other AI agents, and Bank of America said this week it hired 1,000 contractors to fix bugs found by AI.
If the costs of agent misbehavior keep landing on the targets rather than the labs, incentives for companies to police their own models stay weak.