Evidence · DiggingBeagle record
PROTEA: stateful safety evaluation for long-horizon robot tasks
PROTEA: stateful safety evaluation for long-horizon robot tasks
- Source role
- primary disclosure
Evidence record
PROTEA: stateful safety evaluation for long-horizon robot tasks
Claim-level citations (2)
- supportsPROTEA robot hazard memory can over-report risks while tracking delayed effects: PROTEA tests LLM safety judgments on benign and malicious robot plans with harms inserted at different levels of stealthiness; its January 2026 public paper does not establish a rate of prevented real-world injuries.
Abstract, security of task plans and malicious plan dataset
- supportsPROTEA robot hazard memory can over-report risks while tracking delayed effects: The PROTEA paper reports tracking object and task state across robotic actions to detect deferred hazards, while some evaluated LLM judges exhibit materially more false-positive alerts in its benchmark.
Abstract, HarmPlan evaluation and precision/recall by judge