A bigger day from the labs than yesterday: Anthropic shipped its most capable models for daunting technical work, Claude Fable 5.2 and Claude Mythos 5.2 (Mythos, the bug-finding model, is already headed into Cisco and IBM products); Google DeepMind used its new AlphaProof Code to formally verify Signal’s protocol; and Meta introduced Muse Ember, a small model built for the cheap, repetitive work agents do all day. Meta Engineering also published a deep dive on TLX, its hardware-native GPU compiler for production ML. Beyond AI: Git patched a flaw that lets malicious repositories execute code during recursive clones, an SGLang RCE was used to breach NEAR AI and steal data from 18 agents, and Fireworks AI confirmed a wider breach touching customer datasets, model weights and Notion AI chats.
From the labs
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 — Anthropic (+ The Register, Simon Willison)
- Formal Verification of Signal’s Protocol with Gemini — Google DeepMind (AlphaProof Code’s first outing; + The Register)
- Introducing Our Smallest, Most Efficient Model: Muse Ember — Meta (+ Simon Willison)
Worth your time
- How We Built TLX: A Hardware-Native GPU Compiler for Production ML at Scale — Meta Engineering
Beyond AI
- Git security vulnerabilities announced — GitHub (malicious-repo RCE during recursive clone; + The Hacker News)
- SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents and About 100 Users — The Hacker News
- Fireworks AI Confirms Wider Breach Affecting Customer Datasets, Model Weights and Notion AI Chats — The Hacker News (+ The Register)