OWASP: ChatGPT guided a low-skill tester to an admin takeover
Original title: OWASP Finds ChatGPT Helped Low-Skilled User Take Over Admin Account in Hacking Test
On Hack The Box machine Silentium, ChatGPT steered nmap and curl use against a vulnerable Flowise site; autonomous agents did not finish the takeover.
The OWASP GenAI Security Project’s LLM Exploit Generation Phase 2 report says ChatGPT helped an operator with no offensive-security background take over an administrator account in an authorized Hack The Box test. The operator mostly copied commands and returned results; ChatGPT found a hidden staging site running a vulnerable Flowise build, exposed valid usernames, and directed a password change. Autonomous runs with CAI, HexStrike, Claude Code, OpenAI Codex, and hosted or local models researched paths but none completed the final Silentium step, often because provider controls or tool failures stopped them.
Unconfirmed: The cause of one disappeared model response was not confirmed, and the small sample does not measure real-world stealth.
- The persona allowed computer literacy, command-line use, and intermediate prompting.
- Researchers could not reproduce a reply that vanished behind an error.
- Productive hosted runs cost tens of dollars; a local Qwen Coder run still needed corrections.
- OWASP says repeated similar scans of one target may be a detection signal.
- It recommends human approval before high-impact agent actions.
Impact: Administrator account password changed and used to log in
| Product | Affected |
|---|---|
| Flowise | a vulnerable version |
- 1.Require approval before high-impact actions by deployed agents.
- Vendors
- OpenAI
- Products
- FlowiseClaude CodeOpenAI Codex
- Organizations
- OWASPOWASP GenAI Security Project
- AI models
- ChatGPTQwen Coder
From the original article
2 min read at cybersecuritynews.com
OWASP found that ChatGPT helped an operator with no offensive security knowledge take over an administrator account during a controlled hacking test. The operator largely copied commands, returned results to the chatbot, and followed its guidance to change a password and log in. The finding shows how AI can lower the skill needed for some […]
By Abinaya · cybersecuritynews.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.