Check Point Research disclosed a controlled ChatGPT HTML-preview sandbox bypass that could circumvent Lockdown Mode restrictions. Demonstrations distinguish user-dependent phishing or downloads from data exfiltration without a user gesture in ChatGPT Classic. Check Point says OpenAI reported a September 18 fix and that retesting confirmed the technique no longer worked.
Why it matters
It shows how a client rendering boundary can undermine a safety setting, while preserving the reported fix and the demonstrations’ interaction requirements.
Key facts
As stated in the sources, with where to find them.
- The November 5 disclosure concerns a preview-rendering and popup isolation flaw, rather than evidence that the model itself ignored a safety instruction.Technical Details
- The phishing and download demonstrations involve user actions. The ChatGPT Classic exfiltration demonstration works without a user gesture under the tested conditions. These are different scenarios.Attack surface; Zero-Click Exploitation in ChatGPT Classic
- On September 18, OpenAI told Check Point it had deployed a fix; Check Point reports retesting and confirming that the disclosed technique no longer worked.Conclusion
- The source presents researcher demonstrations and responsible disclosure, not an observed in-the-wild campaign.Introduction; Conclusion
Findings that cite this record
No tracked finding cites this record yet.
Key questions this bears on
- Do cyber evaluations of AI agents stay contained?Not reliably. Agents have reached real systems from evaluation, training and research; red-teaming found containment gaps OpenAI says it largely fixed.
Sources
Related records
Oct 29, 2026
Oct 1, 2026
Sep 25, 2026
Sep 16, 2026
Nov 10, 2026
Nov 6, 2026