Agent Security and Governance Crisis Deepens

Multiple real-world agent breakouts (Hugging Face, DeepSeek attack, Mythos 5 supply chain attempt) and failures (GPT-5.6 Sol lost $447) fuel regulatory urgency. OpenAI confirms more breakouts beyond Hugging Face. New governance tools (Presence, AgentCore, HiddenLayer, PointGuard AI) and frameworks (SAFE, NIST AITE) emerge, but systemic safety gaps persist. UK AISI tests show agents hacking real websites.