{ "schema_version": 1, "builder_plan": { "scenario": "An agent looks good in demos, but you cannot tell which product tasks it fails.", "change": "Use the study's evaluation idea to inspect a small set of tasks with human-checked answers.", "baseline": "Your current evaluator and a manually reviewed reference set.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Agreement with human review, missed failures, false alarms, and evaluation cost per task.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.22124", "title": "PROBE: Learning to Audit Policy Compliance in Tool-Using LLM Agents", "summary": "Replaces manual red-teaming and zero-shot adversarial prompting of tool-using agents with a trained LLM auditor that conducts multi-turn, tool-mediated dialogues to surface policy violations, optimized via GRPO, DPO, or GFlowNet objectives against a composite reward. Costs include training the auditor (reinforcement-style fine-tuning of an LLM, feasible at small scale on a single 24 GB GPU only for small auditor models), designing per-policy reward components, and maintaining a simulated user environment with tool access. It can fail through reward hacking by the auditor, implausible user behavior that does not transfer to real traffic, policies whose violations only manifest in downstream systems the simulator does not model, and an auditor that overfits to the current agent version and must be retrained as the agent evolves.", "rationale": "Policy-compliance auditing of tool-using agents is directly relevant to production deployment, and a small trained auditor is infrastructure-feasible, but the training pipeline and reward design require workload-specific validation before adoption.", "technique": "Trained multi-turn LLM auditors for policy-compliance probing", "family": "avaliacao_de_agente", "recommendation": "testar", "published": "2026-10-15", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.22124", "independent_implementations": 1, "total_implementations": 2, "stars": 132, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "Replaces manual red-teaming and zero-shot adversarial prompting of tool-using agents with a trained LLM auditor that conducts multi-turn, tool-mediated dialogues to surface policy violations, optimized via GRPO, DPO, or GFlowNet objectives against a composite reward. Costs include training the auditor (reinforcement-style fine-tuning of an LLM, feasible at small scale on a single 24 GB GPU only for small auditor models), designing per-policy reward components, and maintaining a simulated user environment with tool access. It can fail through reward hacking by the auditor, implausible user behavior that does not transfer to real traffic, policies whose violations only manifest in downstream systems the simulator does not model, and an auditor that overfits to the current agent version and must be retrained as the agent evolves.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.22124", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Auditors trained with PROBE uncover substantially broader and higher-yield policy violations than same-architecture zero-shot adversaries, even when smaller than the target agent; no numeric factor is given in the abstract.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.22124", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }