{
  "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": "2610.26474",
  "title": "AfriGuard: Safety and Robust Refusal For African Languages",
  "summary": "AfriGuard replaces general-purpose or larger off-the-shelf guardrail models for safety classification and refusal detection in African languages with a culturally grounded evaluation dataset plus supervised fine-tuned smaller classifiers. Costs include fine-tuning on the released dataset, maintaining per-deployment classifiers in the inference path (added latency and a failure point), and dependence on dataset coverage of the 10 included languages and safety categories. It can fail on languages or dialects outside the dataset, on culturally specific harm expressions not represented in training data, and because the paper shows models often produce linguistically inadequate responses rather than proper refusals, a detector may report incidental non-assistance as safe behavior.",
  "rationale": "The fine-tuned small classifiers fit constrained infrastructure and the gains are quantified, but the improvement applies only if the reader's production traffic includes the covered African languages and harm categories, so workload-specific validation is required before adoption.",
  "technique": "Culturally grounded multilingual safety guardrails and refusal detection",
  "family": "seguranca_e_guardrails",
  "recommendation": "testar",
  "published": "2026-10-19",
  "as_of": "2026-10-20",
  "editorial_status": "indexed",
  "source_url": "https://arxiv.org/abs/2610.26474",
  "independent_implementations": 0,
  "total_implementations": 0,
  "stars": 0,
  "citations": null,
  "claims": [
    {
      "claim_id": "claim-01",
      "statement": "AfriGuard replaces general-purpose or larger off-the-shelf guardrail models for safety classification and refusal detection in African languages with a culturally grounded evaluation dataset plus supervised fine-tuned smaller classifiers. Costs include fine-tuning on the released dataset, maintaining per-deployment classifiers in the inference path (added latency and a failure point), and dependence on dataset coverage of the 10 included languages and safety categories. It can fail on languages or dialects outside the dataset, on culturally specific harm expressions not represented in training data, and because the paper shows models often produce linguistically inadequate responses rather than proper refusals, a detector may report incidental non-assistance as safe behavior.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.26474",
      "source_page": null,
      "source_excerpt": ""
    },
    {
      "claim_id": "claim-02",
      "statement": "Refusal detectors improve average performance by 43.9% over existing detectors; efficient guardrail models are competitive with models up to 30x larger; base models score below 60% F1 on appropriate safety behavior.",
      "basis": "inferred",
      "result": "",
      "baseline": "",
      "conditions": "",
      "source_url": "https://arxiv.org/abs/2610.26474",
      "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": ""
}