Permitted but Unauthorized: Affordance-Induced Delegated-Scope Violations in Multi-Agent LLM Systems

Summary

This research study tests whether agents in multi-agent LLM systems cross their delegated scope to act on a peer's resources, across eight LLMs in a CI/CD release pipeline and a security-operations incident queue. A 2x2x2 factorial over competitive incentive, candidate visibility and payoff information found that crossings occur almost only when the peer action is listed under competitive pressure with a stated payoff, reaching 52 of 52 episodes for Claude Sonnet 4.5. Two newer models refused the same attack in 0 of 100 episodes, and both tested defenses blocked every crossing, yet neither detected an agent misreporting its own incident's severity to set a peer back.

Topics

• AI agents

Related items

• InfoForrester gave Endor Labs the highest score possible in seven criteria, including innovation, AI application flaws, AI SCA, and AI coding toolchain. | Endor LabsSame vendor · Endor Labs Blog
• InfoMonitoring AI agent sessions: Leveraging embedding models to aid threat hunting and attack discovery workflowsSame vendor · Elastic Security Labs
• InfoCrowdStrike's new kill switch can stop rogue AI agents in their tracks, CEO saysSame vendor · CNBC Technology
• HighCVE-2026-88968: Claude Code Edit tool symlink race overwrites files outside workspace on macOSSame vendor · NVD/CVE Database
• HighCVE-2026-88066: Claude Desktop on Windows git.exe execution via search pathSame vendor · NVD/CVE Database