CONTINUITY: Security-Context Contracts for Composable LLM Agent Controls
The paper’s core claim is that agent security controls can fail at the handoff, even when each control works on its own.
Zheng and Yang call that failure “security-context discontinuity”: context gets dropped, widened, rebound, or reinterpreted as actions move between components. Their CONTINUITY framework uses assume-guarantee contracts and authenticated artifacts to preserve authorization, provenance, policy state, delegation, and execution boundaries through the full instruction-to-effect path. In their reference verifier tests, the full configuration produced no harmful external effects across 2,560 attack instances, while completing 700 benign tasks and escalating 200 ambiguous cases. Source: ArXiv · AI/CL/LG's note.
Zheng and Yang call that failure “security-context discontinuity”: context gets dropped, widened, rebound, or reinterpreted as actions move between components. Their CONTINUITY framework uses assume-guarantee contracts and authenticated artifacts to preserve authorization, provenance, policy state, delegation, and execution boundaries through the full instruction-to-effect path. In their reference verifier tests, the full configuration produced no harmful external effects across 2,560 attack instances, while completing 700 benign tasks and escalating 200 ambiguous cases. Source: ArXiv · AI/CL/LG's note.
score 5