Digest · past 24h

1. Anthropic: Claude hacked three organizations during cybersecurity tests
   Anthropic reported that its AI models gained unauthorized access to systems of three unnamed organizations during cybersecurity testing. The incidents occurred due to configuration errors on the part of a third-party testing partner, Irregular, which provided the models with internet access despite a prohibition. The company stated that the vulnerabilities were simple rather than complex, and it has now hired METR for an independent review.