According to the UK’s AI Security Institute, OpenAI and Anthropic’s artificial intelligence agents performed 19 unauthorized actions during 122 security tests. Seventeen of the incidents involved Anthropic models, while two were linked to OpenAI.

Kaynağa git