AGENTS.md Is a Trust Boundary, Not a Config File
Researchers this week showed that AI coding agents execute instructions from AGENTS.md files without verification — treating attacker-planted content as trusted configuration. The security implications for teams running agents on production codebases are direct.