{ "schema_version": 1, "builder_plan": { "scenario": "An agent reads external documents or tool responses that may contain malicious instructions.", "change": "Replay attacks and legitimate lookalikes in an isolated copy of your agent workflow.", "baseline": "Your existing checks with the same tasks, permissions, and tool responses.", "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": "Successful attacks missed, legitimate actions blocked, detection delay, and review cost.", "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": "2609.19472", "title": "Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models", "summary": "The method replaces an external guardrail model with a small MLP classifier trained on intermediate activations of the deployed LLM (demonstrated on LLaMA-3.1-8B), removing a separate inference pass from the safety pipeline. It costs activation-extraction plumbing inside the serving stack, a per-model labeled training set for the probe, and access to model internals, which excludes API-only models; reported accuracy drops notably on some benchmarks (83-84% F1). It can fail under distribution shift in attack phrasing, when the base model is swapped or fine-tuned (invalidating probe weights), and against adaptive adversaries who learn to produce benign-looking internal states.", "rationale": "For a self-hosted model the approach is cheap to trial and could remove a guardrail inference pass, but the uneven cross-benchmark F1 and dependence on a fixed base model require validation on the reader's own traffic before production adoption.", "technique": "Latent-state MLP probes for harmful prompt detection", "family": "seguranca_e_guardrails", "recommendation": "testar", "published": "2026-09-16", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.19472", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces an external guardrail model with a small MLP classifier trained on intermediate activations of the deployed LLM (demonstrated on LLaMA-3.1-8B), removing a separate inference pass from the safety pipeline. It costs activation-extraction plumbing inside the serving stack, a per-model labeled training set for the probe, and access to model internals, which excludes API-only models; reported accuracy drops notably on some benchmarks (83-84% F1). It can fail under distribution shift in attack phrasing, when the base model is swapped or fine-tuned (invalidating probe weights), and against adaptive adversaries who learn to produce benign-looking internal states.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19472", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "12.6M-parameter probes reach F1 of 99% (WildJailbreak), 83% (Beavertails), and 84% (AEGIS 2.0), reported as competitive with guard models roughly 1000x larger while reducing latency and compute.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19472", "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": "" }