Anthropic's new 'observed exposure' metric ranks 800+ occupations by actual AI usage, not just theoretical risk. Computer programmers top the list at 75%. Unemployment hasn't spiked - but young workers entering exposed fields are finding fewer jobs.

vLLM v0.17.0 adds FlashAttention 4, elastic expert parallelism for live MoE rescaling, full Qwen3.5 support, and a performance-mode flag, all in 699 commits from 272 contributors.

A bipartisan coalition of 40+ groups - from the AFL-CIO to the Congress of Christian Leaders - released a 34-point declaration demanding human control over AI, corporate accountability, and a ban on autonomous lethal weapons.

Alibaba's SWE-CI benchmark tested 18 AI models on 100 real codebases across 233 days of maintenance. Most agents accumulate technical debt and break previously working code. Only Claude Opus stays above 50% zero-regression.

Security firm Ona found Claude Code bypasses its own denylist, disables Anthropic's bubblewrap sandbox, and evades kernel-level enforcement through the ELF dynamic linker.

OpenAI is developing an internal code repository to replace GitHub, putting the company on a collision course with its biggest backer.

Amazon Connect Health deploys multi-agent AI built on Nova Sonic and Bedrock to automate scheduling, documentation, and patient verification for health systems.

Luma AI's new Agents platform, powered by the Uni-1 Unified Intelligence model, lets creative teams go from a written brief to finished video, images, and audio in one workflow.

Arc Institute and NVIDIA release Evo 2, a 40B-parameter open-source AI trained on 9.3 trillion nucleotides from every domain of life, with full weights, code, and training data.

Dario Amodei tells a Morgan Stanley audience he is trying to 'deescalate' the Pentagon dispute, as the FT reports both sides are back at the table with under-secretary Emil Michael.