Search papers, labs, and topics across Lattice.
OpenClaw PRISM is introduced as a zero-fork runtime security layer for tool-augmented LLM agents, addressing vulnerabilities like indirect prompt injection and unsafe tool execution. It integrates a hybrid heuristic-plus-LLM scanning pipeline with conversation-scoped risk accumulation and policy-enforced controls across various lifecycle hooks. Preliminary benchmarks demonstrate its security effectiveness, runtime overhead, and operational recoverability, emphasizing deployable runtime defense for agent gateways.
Stop patching vulnerabilities one-by-one: OpenClaw PRISM offers a defense-in-depth runtime security layer that protects tool-augmented LLM agents across their entire lifecycle.
Tool-augmented LLM agents introduce security risks that extend beyond user-input filtering, including indirect prompt injection through fetched content, unsafe tool execution, credential leakage, and tampering with local control files. We present OpenClaw PRISM, a zero-fork runtime security layer for OpenClaw-based agent gateways. PRISM combines an in-process plugin with optional sidecar services and distributes enforcement across ten lifecycle hooks spanning message ingress, prompt construction, tool execution, tool-result persistence, outbound messaging, sub-agent spawning, and gateway startup. Rather than introducing a novel detection model, PRISM integrates a hybrid heuristic-plus-LLM scanning pipeline, conversation- and session-scoped risk accumulation with TTL-based decay, policy-enforced controls over tools, paths, private networks, domain tiers, and outbound secret patterns, and a tamper-evident audit and operations plane with integrity verification and hot-reloadable policy management. We outline an evaluation methodology and benchmark pipeline for measuring security effectiveness, false positives, layer contribution, runtime overhead, and operational recoverability in an agent-runtime setting, and we report current preliminary benchmark results on curated same-slice experiments and operational microbenchmarks. The system targets deployable runtime defense for real agent gateways rather than benchmark-only detection.
