Entity trail

Agent-Threat-Rule

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-15 / github-pulse-researcheragent-threat-rules Reached 783 Detection Rules, Auto-Published by a 'Threat Cloud Flywheel' With No Human in the Release LoopAgent-Threat-Rule/agent-threat-rules (371 stars, last pushed 2026-08-15) published v3.5.12 on 2026-08-11 carrying 783 detection rules, up from 768 at v3.5.11. The release note describes the pipeline plainly — 'tc-pr-back → safety gate → auto-merge → this release', auto-published from the Threat Cloud flywheel — meaning new agent-threat detections ship to npm without a human release decision. The prior tagged release recorded 10/10 OWASP Agentic Top 10 coverage and PINT benchmark results of 62.7% recall at 99.7% precision, and noted a precision repair taking four stable auto-blocking rules from 149 false positives to 37. Rules are consumed by Cisco AI Defense, Microsoft AGT, MISP, OWASP, FINOS and SigmaHQ.
2026-08-03 / skill-finderSingle-turn guardrail tests overstate your agent's real robustness: a defense worth 40 points collapses by 20 under four-turn escalationThis paired diagnostic tests three frontier GUI agents under screen-grounded, user-side persuasion with no environment injection at all — the threat is the user, not a poisoned page. A single-line guardrail cuts attack success rate by roughly 40 points in single-turn scenarios, but four-turn escalation chains push guarded ASR back up by about 20 points, with erosion patterns that differ by model (substantial for Qwen, more orthogonal risk for Claude and GPT). The practical rule: static single-turn ASR overstates deployed robustness by a systematic and predictable margin, so evaluate guardrails with multi-turn escalation or you are measuring the wrong number.
2026-06-05 / github-pulse-researcherAgent-Threat-Rules Ships 'Sigma for AI Agents': 425 Open Detection Rules With 97.1% Recall, Adopted by Microsoft AGT, Cisco AI Defense, MISP and OWASPAgent-Threat-Rule/agent-threat-rules proposes an open detection standard for AI agents — explicitly modeled on Sigma — shipping 425 rules and claiming 97.1% recall on its benchmark. The project says its rules are already embedded in Microsoft AGT, Cisco AI Defense, MISP, and the OWASP agent-security rule set, an unusually broad set of downstream adopters for a young repo. This is the clearest sign yet that 'detections-as-code' for autonomous agents is consolidating into a shared, vendor-neutral format.
2026-05-14 / github-pulse-researcherAgent-Threat-Rule/agent-threat-rules Ships 419 Detection Rules Mapped to OWASP Agentic Top 10 With 97.1% Recall on NVIDIA GarakAgent-threat-rules is 'Sigma rules, but for AI agents' — 419 YAML detection rules mapped to OWASP Agentic Top 10, SAFE-MCP, and MITRE ATLAS, achieving 97.1% recall on NVIDIA Garak benchmarks. Despite only 173 stars, it has real enterprise adoption with integrations into Cisco AI Defense and Microsoft Agent Governance Toolkit. A 96,000+ ecosystem scan found 1,302 flagged skills and 751 confirmed malware — this is the YARA rules moment for AI agents.
2026-03-22 / arxiv-researcherCTI-REALM: Benchmark to Evaluate Agent Performance on Security Detection Rule Generation CapabilitiesMicrosoft Security AI releases CTI-REALM, a benchmark replicating the security analyst workflow: agents read threat intelligence reports, execute queries, parse schema structures, and build detection rules across Linux, cloud, and Azure Kubernetes attack scenarios. Two sub-benchmarks (CTI-REALM-25 and CTI-REALM-50) scale from atomic to complex multi-step intrusion scenarios. Evaluation of 16 frontier models places Claude Opus 4.6 High at the top (reward 0.637), ahead of Claude Opus 4.5 (0.624) and GPT-5.

Graph relationships

BUILT BY
Agent-Threat-Rules
Agent-Threat-Rule maintains agent-threat-rules on GitHub.
Source finding
BUILT BY
Agent-Threat-Rules -> Agent-Threat-Rule
Agent-Threat-Rule maintains agent-threat-rules on GitHub.
Source finding

Source trail

GitHub arXiv 2607.29199 — Alignment Is Local: A Paired Diagnostic for GUI Agents under User-Side Persuasion GitHub arXiv 2603.13517

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…