OWASP tests show guided AI can help novices finish a lab takeover
First merged summary.
OWASP’s October 27 guide found a guided ChatGPT session completed an authorized Hack The Box admin takeover, while autonomous agents did not.
OWASP’s GenAI Security Project published the LLM Exploit Generation Guide 2.0, or Phase 2 report, on October 27. It compared human-guided models with autonomous agents on Hack The Box machines Silentium and MonitorsFour, released after the models’ training cutoffs. According to the report, ChatGPT steered an operator with no offensive-security background through nmap and curl commands against a vulnerable Flowise staging site, exposing valid usernames and directing a password change that took over an administrator account. Stronger agents using CAI, HexStrike AI, Claude Code, and OpenAI Codex reached the account-takeover stage on Silentium in some of nine configurations, but none reliably completed the final step because of provider guardrails or tool failures. OWASP treats repeated similar scans of one target as a possible detection signal and recommends limited permissions, treating external content as untrusted, and human approval before high-impact actions.
- Tests covered Silentium and MonitorsFour, released after the models’ cutoffs.
- ChatGPT found a hidden staging site running a vulnerable Flowise build.
- Autonomous runs used CAI, HexStrike AI, Claude Code, and OpenAI Codex.
- Productive hosted runs cost tens of dollars; local Qwen Coder still needed corrections.
- Researchers could not reproduce one reply that vanished behind an error.
- OWASP recommends approval before sensitive or high-impact agent actions.
Timeline
dated events in the reporting · oldest first- Oct 27, 2026OWASP publishes LLM Exploit Generation Guide 2.0. · GBHackers
What each outlet adds
beyond the shared facts- GBHackersNames both machines, nine configurations, and the October 27 publication.
- Cyber Security NewsDetails Flowise, ChatGPT’s command guidance, costs, and the vanished reply.
- Organizations
- OWASPOWASP GenAI Security Project
Coverage
oldest first · one row per article · UTC- · 2d ago
GBHackersFirst report
OWASP finds AI agents lower the skill bar for parts of hacking
Guide 2.0 tested guided models and agents on two Hack The Box machines released after the models’ cutoffs.
- · 2d ago
Cyber Security News
OWASP: ChatGPT guided a low-skill tester to an admin takeover
On Hack The Box machine Silentium, ChatGPT steered nmap and curl use against a vulnerable Flowise site; autonomous agents did not finish the takeover.