The Collapse of Trust: Security Architecture for the Age of Autonomous AI Agents (v4)

v4 Update (April 2026) — Three Papers This release contains three research papers on AI agent security: The Collapse of Trust (v4, 23 pages, 65 references) — The complete ATR thesis. Updated from v3 with 108 rules (was 61), dual benchmarks (PINT 99.6% precision + SKILL.md 96.9% recall), 53,577-skill mega scan, and Cisco AI Defense adoption. 53,577 Skills, 946 Threats (10 pages, 25 references) — The largest AI agent security scan ever conducted. 53,577 skills across OpenClaw and Skills.sh. 946 flagged, 875 critical, 0% false positive. Tool description poisoning accounts for 71% of detections. 30 CVEs in 60 Days (14 pages, 22 references) — Empirical analysis of the MCP attack surface. 30 CVEs in 60 days (15x faster than Docker). 38% of MCP servers have zero authentication. 7-class attack taxonomy. Key findings: 108 ATR detection rules across 9 threat categories PINT benchmark: 99.6% precision, 61.4% recall (850 adversarial samples) SKILL.md benchmark: 96.9% recall, 100% precision, 0% FP (498 real-world samples) 53,577 skills scanned, 946 threats found (1.77% flagged rate) Adopted by Cisco AI Defense as upstream detection rules OWASP Agentic Top 10: 10/10 coverage Open source: github.com/Agent-Threat-Rule/agent-threat-rules