{
  "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": "2610.31376",
  "title": "Certified Execution Harness for Agentic AI: Safe and Least-Intrusive Petri-Net Supervisory Control Between Reasoning and Action",
  "summary": "The method replaces prompt-wording-dependent guardrails (filters, judge models, system-prompt instructions) with a formal supervisory layer between the LLM and its tools, where Petri-net marks represent capabilities, approvals, and budgets so that any request lacking unspent preconditions is refused regardless of phrasing. It costs the engineering effort of formally modeling each tool's preconditions and budget semantics, plus one inference call per replanning retry after a refusal; no additional GPU or model training is required. It can fail if any of the four stated transfer assumptions is violated in deployment, if the capability model under-specifies a tool's side effects, and the paper's key empirical finding is that safe release pipelines require a supervisor-refusable step after rollout starts, which may conflict with existing CI/CD designs.",
  "rationale": "A formally enforced capability-and-budget gate is implementable on constrained infrastructure without GPU cost, but its safety guarantees depend on hand-built models of each tool's preconditions that must be validated per workload before production reliance.",
  "technique": "Petri-net supervisory control for agent tool execution",
  "family": "seguranca_e_guardrails",
  "recommendation": "testar",
  "published": "2026-10-22",
  "as_of": "2026-10-23",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.31376",
  "independent_implementations": 0,
  "total_implementations": 1,
  "stars": 10,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "The method replaces prompt-wording-dependent guardrails (filters, judge models, system-prompt instructions) with a formal supervisory layer between the LLM and its tools, where Petri-net marks represent capabilities, approvals, and budgets so that any request lacking unspent preconditions is refused regardless of phrasing. It costs the engineering effort of formally modeling each tool's preconditions and budget semantics, plus one inference call per replanning retry after a refusal; no additional GPU or model training is required. It can fail if any of the four stated transfer assumptions is violated in deployment, if the capability model under-specifies a tool's side effects, and the paper's key empirical finding is that safe release pipelines require a supervisor-refusable step after rollout starts, which may conflict with existing CI/CD designs.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.31376",
      "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": ""
}