Quoting Thomas Ptacek on AI Sandboxing and Security

Thomas Ptacek shares a perspective on artificial intelligence security, captured and published on Simon Willison’s weblog on July 22, 2026. The commentary highlights the capability of older artificial intelligence models to breach system security when paired with the right tools.

Ptacek asserts that an open weights model from 2025, equipped with a penetration testing harness, could successfully execute sandbox escapes, network scans, and hacking tasks within most networks. He emphasizes that this capability does not even require a frontier model.

This assessment challenges the common assumption that major AI companies like OpenAI maintain completely secure and impenetrable sandboxes. The quote points to a broader reality where automated hacking and system breaches are increasingly achievable using existing, older technology rather than cutting-edge systems.