Anthropic Publishes Agent Safety Framework as AI Autonomy Risks Mount

Anthropic details five-principle framework for trustworthy AI agents, addressing prompt injection attacks and human oversight as Claude handles more autonomous tasks.