Modified ARTEX agent pentest framework stripped of programmatic scope enforcement
AccBalanced · x · 2026-11-21
CyhawkAfrica's review of a modified version of ARTEX — an LLM multi-agent autonomous pentest framework (Go + Next.js + SQLite) combining AI task orchestration, recon, vuln research, browser interaction and command execution — found changes that go beyond new capabilities. The modified version removed programmatic Rules-of-Engagement controls: global asset blocklists, task-level allow/deny rules, and authorization checks before assets are added. Remaining natural-language constraints are passed only as prompt instructions to the agents, offering no technical enforcement against out-of-scope targets. The findings raise security concerns about controls around agentic offensive-security tools.
More from coding & agent
- Unsloth release runs 26.4B Inkling 2 Mini on 16GB RAM, cuts MCP token use 87% — danielhanchen · 2026-11-21
- Vibe-coding culture: everyone builds over the holidays, few realize when to log off — haydendevs · 2026-11-21
- dotey: Agents Don't Need Full Codebase Understanding — Verify Instead — dotey · 2026-11-21
- Agent learns continually via knowledge-graph persistent memory of user history — khademinori · 2026-11-21
- AFK Pilot VS Code extension adds Gemini 4 Neon, multi-harness steering, 180K installs — PawelHuryn · 2026-11-21
- Dev compiles Rails into the browser for offline, millisecond-response apps — kieranklaassen · 2026-11-21