AnalysisCybersecurityJuly 22, 2026
Ptacek: 2025 open weights model could sandbox escape and hack networks
Thomas Ptacek states that an open weights model from 2025 with a pentest harness could escape sandboxes and compromise most networks. He adds that this is only surprising given OpenAI's unsound sandboxes.
Featured · Thomas Ptacek
1 source
More stories today
- ServiceNow invests $40M in Indian banking AI firm BusinessNext
- 12 of top 50 US entertainment apps are AI-generated short-form dramas
- STMicro forecasts revenue growth as AI boom boosts chipmaker
- Atomic runs AI agents entirely on your local machine
- @mehul's cleaning robot launches after six-year development