AnalysisCybersecurityJuly 22, 2026
Ptacek: Open weights model from 2025 could escape and hack most networks
Thomas Ptacek suggests an open-weights model from 2025 with a pentest harness could perform sandbox escapes and scan/hack most networks, challenging assumptions about OpenAI's sandboxing.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Pro tip: distill harnesses alongside models
- Yahoo enhances search retargeting with Amazon Bedrock
- AWS introduces inference meta-monitoring for SageMaker AI endpoints with Amazon Quick
- Okta acquires AI security startup Permiso for about $200M
- EvoLib lets LLMs learn from their own inference experience