Papers.

AI agents retrieve data, call APIs, send messages, and trigger workflows across systems. Every action involves delegation, but four commonly conflated concepts (authority, access, on-behalf-of, identity) answer different questions and carry different risks. This paper provides a formal comparison, maps threat vectors to delegation failures, and proposes a zero-trust agent framework separating mandate from identity from capability from execution.

Enterprises are beginning to use AI assistants and agentic systems as reusable work environments. Employees create prompts, workflows, helper scripts, and small pieces of automation, then share them as "skills". This paper explores why skills create a different risk model from traditional software, and proposes a governance framework combining software lifecycle discipline with AI-specific runtime enforcement.