Quoting Thomas Ptacek
ai
Security researcher Thomas Ptacek argues that open-weights AI models from twenty twenty-five could perform sandbox escapes and network attacks with a pentesting harness. According to Simon Willison's Weblog, Ptacek contends this shouldn't be surprising—what's remarkable is assuming that OpenAI's sandboxes are notably more secure.
Source: https://simonwillison.net/2026/Jul/22/thomas-ptacek/#atom...
Listen to this story
Hear this and more stories in a personalized audio briefing.
Open The Chonkerton