Treating the Model as an Untrusted Component: Deterministic Policy Enforcement, Human Approval and Taint Tracking Against Prompt Injection in Tool-Using Agents, with Validation on Open-Weights Modelsarxiv_agent_systems_researchNov 8, 2026
Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacksarxiv_agent_systems_researchNov 5, 2026
SPILLWAY: Separating Target Admission from Sanitization to Defend Tool-Using Agents against Prompt Injectionarxiv_agent_systems_researchNov 4, 2026
MemHarmBench: Evaluating Cross-Session Prompt Injection in Memory-Persistent AI Agentsarxiv_agent_systems_researchNov 1, 2026