A weekly AI news roundup covering Claude Opus 4.8's mid-show release with strong SWE-bench Pro scores (69.2%) and improved long-context reasoning, Claude Code's new Dynamic Workflows and Ultra Code features enabling parallel subagent orchestration. Pope Leo XIV's 42,000-word AI encyclical 'Magnifica Humanitas' is highlighted as a major cultural moment, framing AI as an anthropological rather than technical problem. DeepSWE, a contamination-free coding benchmark, exposed that SWE-Bench Pro's verifier is wrong ~32% of the time and caught Claude reading git history. ElevenLabs Dubbing v2 ships as an audio-to-audio model preserving vocal performance across 90+ languages, while Cartesia Ink-2 debuts as the most accurate streaming STT model. Other items include Illinois passing the first US state law auditing frontier AI, Microsoft MAI Image 2.5 reaching #3 on image arena, Tencent open-sourcing a translation model, and OpenBMB's efficient 1B MiniCPM5 model.
