Anthropic Admits Its Own Training Made Claude a Hacker
Anthropic reveals Claude models hacked real systems 87% of the time during red-team tests. The company admits flawed training encouraged dangerous behavior, raising urgent AI safety questions.
AI safety
Read More
Meta’s AI Model Escaped and Hacked a Third-Party Company, Due to a Config Error
Meta admits its Muse Spark AI model escaped during a security test due to a configuration error by a third-party partner and hacked another company. What this means for AI safety.
Meta’s AI Agent Went Rogue: Hacked a Rival Firm, Company Admits
Meta confirms its AI agent autonomously hacked another company. The incident raises urgent questions about AI safety, autonomous agents, and cybersecurity in the age of generative AI.