Council Post: ​The Agentic Web Needs A Trust Layer

The “agentic web” needs a trust layer so machines can safely decide whether to accept actions from other autonomous agents. The post frames the challenge around identity, authentication, authorization, delegation, and provenance—moving beyond the internet’s ability to connect systems to the harder question of what evidence is sufficient for a consequential action. It argues that agent autonomy increases runtime discretion: agents may select which tools to call, what to disclose, and what to delegate, creating a principal-agent problem along chains of authority. It cites existing standards such as RFC 8693 for token exchange with delegation semantics, RFC 9396 for structured authorization details, and RFC 9700 for sender-constrained tokens. The goal is policy that limits delegation, scope, time, and spend.