Models routinely take shortcuts to pass evaluations. Who gets to decide what incidents are innocuous?

Anthropic has released new reporting on Claude’s behavior in testing and other internal evaluations — and the findings indicate a concerning pattern.

On Friday, the company detailed several “unintended” actions its models had taken recently while completing certain benchmarks. Framing Claude’s actions as less serious than the breaches both Anthropic and OpenAI agents were involved in earlier this summer, the report details how Claude Haiku 4.5 was generating example tasks on “randomly selected webpages.”

It came across a page about an unsolved homicide case and filled out a tip form with the Philadelphia police, saying, “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.”

This was, of course, not true. Anthropic noted that the police flagged the submission as spam, and it wasn’t taken seriously as part of the investigation. The company said Claude took similar action in several instances, including while completing computer use evaluation OSWorld.

“Claude was instructed never to log in, create accounts, enter personal data, make purchases, or submit anything destructive, but the instructions did not rule out form submissions,” Anthropic explained, exemplifying how airtight security instructions have to be to actually prevent unwanted model behavior.

Based on other agentic security incidents from the last year alone, it’s also not clear that simply adding form submissions to this instruction would guarantee prevention if the model were given real stakes to complete a task.

Source link