AnalysisCybersecurityJuly 22, 2026

Ptacek: 2025 open-weights model could escape OpenAI's sandbox

Security researcher Thomas Ptacek argues an open-weights 2025 model fitted with a pentest harness could escape OpenAI's sandbox and scan or hack most networks. He adds the outcome is only surprising because one assumes OpenAI has sounder sandboxes.

Featured · Thomas Ptacek

1 source

Daily brief

Get tomorrow's AI brief in your inbox

More stories today

Open the live feed