A new red-teaming framework walked into production AI agents -- the coding assistants wired into your files and inbox -- and got in almost every time, with a chilling twist: the better an agent is at its job, the easier it is to break. We unpack how it grades what agents actually do instead of what they claim, then the two defenses fighting back -- auditing the tool-and-skill supply chain, and fast safety guards that reason in training but not at runtime. Plus the elegant flip side: teaching a model to page its own memory like an operating system. And in the news: Meta's Muse models, the slop cleanup economy, and a copyright reframe that could touch every AI company.

Chapters

0:00 Cold Open -- The Machine That Breaks Agents
0:50 The Headlines -- Meta's Muse, the Slop Cleanup Bill, and a Copyright Reframe
6:09 Intro -- Cheap to Make, Expensive to Trust
7:12 The Crack -- Testing What Agents Do, Not What They Say
11:41 The Door -- Auditing the Plumbing, Not the Prompt
16:07 The Memory -- Paging an AI's Mind Like an Operating System
20:51 Wrap-Up -- Trust Is the Product Now

Links

Cold Open -- The Machine That Breaks Agents -- https://arxiv.org/abs/2607.01793
The Door -- Auditing the Plumbing, Not the Prompt -- https://arxiv.org/abs/2606.31227
The Memory -- Paging an AI's Mind Like an Operating System -- https://arxiv.org/abs/2606.31145