Third-party dispatches, not graded receipts
Anthropic details unintended Claude agent actions, including 20 visa form submissions
Anthropic published a report on 'unintended model actions' observed in evaluations and internal use, including a false homicide tip to Philadelphia police and incomplete visa applications submitted via a State Department website. The company says it is cutting internal evals off from the live internet in response.
“The actions [Anthropic observed](https://www.anthropic.com/research/investigating-unintended-model-actions) from its Claude AI include “Claude submitting a sensitive form on a real website when it should not have,” and the company detailed how Claude gave Philadelphia police [a fake tip about an unsolved homicide](https://www.theverge.com/ai-artificial-intelligence/1009090/anthropic-fake-homicide-information-philadelphia-pd-tip).”
Bears on: altman-agents-join-workforce-2025