Anthropic discloses Claude now writes 80% of its own codebase and delivered a 52× training speedup — with human judgment the last narrowing frontier.
Calif used 'Mythos' AI tooling to build a macOS kernel exploit bypassing Apple's Memory Integrity Enforcement — LLM-assisted offensive security has reached kernel-level exploitation.
Anthropic's 1M-conversation study on Claude sycophancy drove training changes in Opus 4.7 and Mythos Preview, cutting the rate by up to 75%.
The White House is blocking Anthropic's Mythos expansion to 70 organizations, exposing a critical gap: AI cybersecurity risk is entirely self-reported.
Anthropic restricted Mythos via self-assessment, but OpenAI and Google face no external governance at the same threshold — cyberrisk assessment is entirely self-reported.
Anthropic's Mythos — flagged as too dangerous to release — was accessed by guessing its endpoint URL after naming conventions leaked via the Mercor data breach, exposing a critical API security gap.
Curated AI insights — sent when there's something worth your inbox.