---
title: "Thomas Ptacek: 2025 Open-weights Models Could Perform Sandbox Escape and Pentesting"
source: "Simon Willison 博客（Atom）"
source_tier: "t2"
url: "https://simonwillison.net/2026/Jul/22/thomas-ptacek/#atom-everything"
permalink: "https://feed.astralor.com/items/itm_367bb2eccf8c6e882da2c29a"
published: "2026-07-22T23:59:01.000Z"
score: 56
channel: "ai"
tags: ["AI Security","Thomas Ptacek","Open-weight models","Sandbox Escape","Penetration testing"]
---

# Thomas Ptacek: 2025 Open-weights Models Could Perform Sandbox Escape and Pentesting

## AI commentary

> As an editor, I recommend security practitioners pay attention to this dynamic: Ptacek captures in one sentence the real threat that model capability leaps could bring, making it worth considering next year's upgrade roadmap for your red team toolkit.

## At a glance

Security expert Thomas Ptacek argues that if a pentest harness were built for an open-weights model from 2025, it could perform sandbox escapes, scanning, and hacking. This highlights the potential disruptive power and practical value of open models in cybersecurity, sparking new considerations about model misuse.

## Key points

- Ptacek believes 2025 open-weights models can perform sandbox escapes
- With a dedicated pentest harness, they can automate scanning and attacks
- The view implies open-model security risks are moving from theory to practice

## Full text

I genuinely believe that if you took an open weights model from 2025 and built a pentest harness for it, it could do this kind of sandbox escape and scan/hack …

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_367bb2eccf8c6e882da2c29a