{ "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": "2610.00977", "title": "ABSENTIA: Detecting Broken Access Control Vulnerabilities in Web Applications", "summary": "ABSENTIA replaces undirected LLM-agent code review and rule-based static analyzers (CodeQL, Semgrep) with a structured audit loop: agents build a route-to-code graph, then apply invariant falsification route by route, inferring intended authorization properties and reporting unenforced ones for maintainer review. Cost is API inference proportional to the number of routes (graph construction plus per-route analysis and verification), plus human triage of reports; no local model or cluster is needed, so a 24 GB GPU or third-party API budget suffices, but total token spend scales with codebase size and must be measured on the reader's repositories. Failure modes are material: only 51% of findings pass LLM verification, so roughly half of reports may be false positives consuming maintainer time, recall is 63% at best, and results are validated on only 30 recent advisories across 9 frameworks, leaving generalization to other stacks unproven.", "rationale": "It runs on API-scale infrastructure and substantially outperforms both static analyzers and unstructured agents on a human-verified benchmark, but the 51% verifier confirmation rate and small benchmark mean teams must validate precision and token cost on their own codebase before relying on it in production audits.", "technique": "LLM-agent scaffolding for access-control vulnerability audits", "family": "seguranca_e_guardrails", "recommendation": "testar", "published": "2026-10-01", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.00977", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "ABSENTIA replaces undirected LLM-agent code review and rule-based static analyzers (CodeQL, Semgrep) with a structured audit loop: agents build a route-to-code graph, then apply invariant falsification route by route, inferring intended authorization properties and reporting unenforced ones for maintainer review. Cost is API inference proportional to the number of routes (graph construction plus per-route analysis and verification), plus human triage of reports; no local model or cluster is needed, so a 24 GB GPU or third-party API budget suffices, but total token spend scales with codebase size and must be measured on the reader's repositories. Failure modes are material: only 51% of findings pass LLM verification, so roughly half of reports may be false positives consuming maintainer time, recall is 63% at best, and results are validated on only 30 recent advisories across 9 frameworks, leaving generalization to other stacks unproven.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00977", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Recalls 19 of 30 verified broken-access-control advisories (17 under paired credit) versus 3 for an unstructured agent on the same model, while CodeQL and Semgrep recall none; an LLM verifier confirms 51% of findings.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.00977", "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": "" }