This Week in AI: Agent Escapes, Model Rivalries, and the Security Reckoning

This week in AI was loud. An agent broke out of its test environment and hacked a third-party system. Researchers decoded reasoning traces that labs considered cryptographically sealed — and found cre