ARTEX, built by a university student, tames autonomous security-testing agents with dual-graph stateful design
cyb3rops · x · 2026-11-20
Security researcher cyb3rops boosted ARTEX, a project created by a young university student.
Slide content from the talk shows ARTEX is an autonomous agent architecture for security testing, built around the principles "autonomy ≠ no human oversight" and "finding a vulnerability ≠ auto-reporting", with a Stateful design:
- Dual-graph architecture: exploration graph + asset graph, quantifiable testing progress and clearer task goals
- Controllable: a Planner corrects/stops running Workers; humans can inject new intents, goals, and operation constraints in real time
- Rules-first: rules-first (batched) plus low-cost models in a layered strategy
- Stable long tasks: model chains configured with automatic failover on single-point failure
- Traceable: every model/tool call and outbound request is fully logged and replayable
A useful reference for anyone building pentest or vuln-discovery agents: human-in-the-loop plus observability-first engineering.
More from coding & agent
- Coding Bake-off: Hy4 Flash Builds Subway Surfers Clone in 7m38s, DeepSeek-V4.1-Flash Takes 90 Minutes — aliscodes · 2026-11-20
- Vork: open-source headless 3D and VFX engine built for AI agents — ravisparikh · 2026-11-20
- No Slop Manifesto tops 1,000 signatures: humans review the plan, design, result — annbordetsky · 2026-11-20
- Victor Eijkhout's free Parallel Programming textbook now online with AI tutor — burkov · 2026-11-20
- OrcaRouter launches Shield: AI-native defense that deceives offensive agents with decoys — cyb3rops · 2026-11-20
- The 6-folder second brain: a schema doc so Claude stops re-learning your business — evielync · 2026-11-20