{ "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": "2609.16098", "title": "Universal Defenses for Tool-Integrated LLM Agents Against Adversarial Attacks", "summary": "The method replaces undefended agent execution with a layered defense stack: Isolation Forest anomaly detection to filter attacker-injected tools, restoration of the original toolset before planning, plus Chain-of-Thought prompting, self-reflection, and task paraphrasing. It costs additional inference passes for reasoning and paraphrasing (raising latency and API spend), a white-box toolset snapshot requirement, and an anomaly detector that needs tuning against false positives on legitimate tools. It can fail against adaptive attacks crafted to evade the anomaly detector, against novel attack vectors outside the four studied categories, and through over-filtering that removes tools needed for task completion.", "rationale": "The defenses are cheap, modular, and API-compatible, but the 0% ASR claim is benchmark-specific, so the reader should validate detection thresholds and robustness against adaptive attacks on their own tool distribution before production deployment.", "technique": "Tool filtering and prompt-based injection defenses", "family": "seguranca_e_guardrails", "recommendation": "testar", "published": "2026-09-14", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.16098", "independent_implementations": 2, "total_implementations": 2, "stars": 3, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces undefended agent execution with a layered defense stack: Isolation Forest anomaly detection to filter attacker-injected tools, restoration of the original toolset before planning, plus Chain-of-Thought prompting, self-reflection, and task paraphrasing. It costs additional inference passes for reasoning and paraphrasing (raising latency and API spend), a white-box toolset snapshot requirement, and an anomaly detector that needs tuning against false positives on legitimate tools. It can fail against adaptive attacks crafted to evade the anomaly detector, against novel attack vectors outside the four studied categories, and through over-filtering that removes tools needed for task completion.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.16098", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Reduces Attack Success Rate to 0% in many settings across direct/indirect prompt injection, memory poisoning, and backdoor attacks, while preserving or improving task success rate.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.16098", "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": [ { "anchor": "S3.EGx1", "label": "", "section": "3 Methods", "role": "proposed_method", "latex": "y_{i}=\\mathcal{F}_{\\text{IF}}(\\phi(\\tau_{i}))=\\begin{cases}1&\\text{if }\\tau_{i}\\text{ is an outlier}\\\\\n0&\\text{otherwise}\\end{cases}", "mathml": "yi=ℱIF​(ϕ⁡(τi))={1if ​τi​ is an outlier0otherwise", "context": "Each tool is labeled as normal or anomalous:", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "A1.Ex2", "label": "", "section": "Appendix A ATF Algorithm", "role": "proposed_method", "latex": "y_{i}=\\mathcal{F}_{\\text{IF}}(\\phi(\\tau_{i}))=\\begin{cases}1&\\text{(outlier)}\\\\\n0&\\text{(normal)}\\end{cases}", "mathml": "yi=ℱIF​(ϕ⁡(τi))={1(outlier)0(normal)", "context": "This project was supported by research funding from Canadian Artificial Intelligence Safety Institute.", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-10-11", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }