https://theknowledge.substack.com/s/ananke-briefshttps://substackcdn.com/image/fetch/$s_!lWh0!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F414e4adc-b8dc-4838-9245-3fefe79671f6_1280x1280.pnghttps://theknowledge.substack.com/s/ananke-briefsSubstackSun, 23 Aug 2026 18:53:37 GMThttps://theknowledge.substack.com/p/openais-model-escaped-its-sandboxhttps://theknowledge.substack.com/p/openais-model-escaped-its-sandboxSat, 25 Jul 2026 12:37:12 GMTOpenAI gave two of its latest models a cyber test. They found a way out of the sandbox, got access to the internet, hacked an entirely different company (Hugging Face) and tried to steal the answers.

Hugging Face then tried using commercial AI models to investigate the attack, but their safeguards blocked the evidence. It ended up using a Chinese open model (GLM) instead.

I talk through the safety failure, the incentives pushing the frontier labs to keep racing, and the growing fight over Chinese models and distillation.

00:00 AI Frontiers: Breakouts, Distillations, and Rivalries
00:59 Crash Test Dummies
24:38 The Distillation Wars

]]>https://theknowledge.substack.com/p/frontier-ai-compute-control-and-chinahttps://theknowledge.substack.com/p/frontier-ai-compute-control-and-chinaFri, 24 Jul 2026 13:34:49 GMTMoonshot released Kimi K3 on Thursday and had to stop taking new subscribers by Sunday because it did not have enough compute to meet the demand.

I talk about what that says about China’s progress despite US chip controls, the difference between releasing a model and having the infrastructure to run it, and the growing role governments are playing in deciding who gets to build and use frontier AI.

00:00 Frontier AI: Compute, Control, and China
01:34 Export Controls
10:16 The Fable Shutdown
14:52 Compute Constraints

]]>