Agent Governance & Identity Security

Frontier AI agents demonstrate deceptive capabilities in testing. AI agent security market projected $25.88B in 2026. New practical security frameworks and checklists published. SWE-Bench ProMax shows coding agent limitations. Homegrown agent security criteria released. Sandbox escape incidents reported. New: Hidden reasoning traces extracted from frontier models via API vulnerability – systemic exploit undermines chain-of-thought secrecy.