AI News Briefing — May 16, 2026

Zyphra released ZAYA1-8B-Diffusion-Preview, the first Mixture-of-Experts diffusion model converted from an autoregressive LLM, achieving up to 7.7x inference speedup with no systematic loss in evaluation performance. This demonstrates a viable path to much faster LLM inference through diffusion-based decoding.

Greg Brockman has officially taken control of OpenAI's products as part of a reorganization effort to unify ChatGPT and Codex into one core product experience. This represents another significant executive shake-up at the company.

A federal judge delayed final approval of Anthropic's $1.5 billion copyright settlement—the largest in US history—after authors objected to high lawyer compensation and low class member payouts. The case has significant implications for how AI companies resolve training data disputes.

OpenAI is exploring legal options against Apple after the ChatGPT-Siri integration allegedly failed to deliver expected subscription revenue. OpenAI suspects Apple intentionally under-promoted the integration, potentially damaging the ChatGPT brand.

Poetiq's Meta-System achieved new state-of-the-art on LiveCodeBench Pro by automatically building and optimizing its own inference harness without fine-tuning. It boosted GPT 5.5 High from 89.6% to 93.9% and Gemini 3.1 Pro from 78.6% to 90.9%, surpassing Google's Gemini 3 Deep Think.

Continuing our coverage from yesterday, The Musk v. Altman trial has reached closing arguments, with a federal jury now deciding the outcome. The trial has reportedly made both sides look bad through damaging revelations.

Continuing our coverage from yesterday, Mira Murati, former OpenAI CTO, shared her vision for Thinking Machines Lab, emphasizing AI that collaborates with humans rather than automating them away. She explicitly positioned the company against full automation approaches.