AI Industry Daily News
A roundup of the AI industry's day, centered on Codex Windows support, grok-build-0.1, Claude Opus 4.8, Command A+, and Rosalind Biodefense.
Today's highlights
WSJ Opinion Criticizes Newsom's AI Workforce Executive Order, Urges Inaction
A Wall Street Journal opinion piece by Barton Swaim argued state governments should avoid large interventions over AI-driven labor disruption, targeting California Governor Gavin Newsom's recent executive order.
Key topics and reactions
WSJ Opinion Criticizes Newsom's AI Workforce Executive Order, Urges Inaction
Newsom signed Executive Order N-6-26 on May 21, 2026, directing state agencies to prepare for AI's economic impact on jobs and small businesses. The order calls for research and data collection on early-warning signals, a review of the WARN Act within 180 days, and study of worker-support measures including severance, equity compensation, employee-owned structures, and universal basic capital, alongside expanded job training. It carries limited legal force but includes deadlines such as an October 15, 2026 report.
Newsom has framed the measure as an 'AI New Deal,' comparing it to FDR's social contract, while emphasizing California as an AI innovation hub home to 33 of the world's top 50 AI companies. The state cites concern over disproportionate effects on women and other groups.
Swaim countered that prior large interventions such as the Green New Deal and Build Back Better failed to deliver results, and that the AI-driven crisis has not yet materialized.
Web Inventor Berners-Lee Calls for User Control of Personal Data in AI Design
In a conversation moderated by Bloomberg's Francine Lacqua and Tom Mackenzie, Berners-Lee described AI as exciting but argued companies should avoid unrestricted data access and ensure individuals can control their own data. The view is consistent with remarks he made at SXSW London earlier in June.
He argued the web's original principles of openness, universality, and primacy of the individual should apply to AI, and noted concern that generative AI models train on vast amounts of web data. He also said the AI field lacks an international standards body comparable to the W3C, which set HTML and HTTP standards.
Berners-Lee's position connects to Inrupt, the company he co-founded in 2018, which promotes a decentralized architecture storing user data in secure 'Solid pods.' Inrupt is developing an AI assistant called Charlie that filters queries to minimize personal information shared with external tools such as ChatGPT and Claude.
OpenAI Releases GPT-5.5 Instant With Improved Health-Question Performance
The model is OpenAI's flagship fast model used by 230 million people weekly, and the company emphasized safety improvements at that scale, including better recognition of cases that need urgent attention.
Separately, OpenAI reported diagnostic results on rare pediatric diseases using o3 Deep Research, in collaboration with researchers at Boston Children's Hospital and Harvard, published in NEJM AI.
The applications arrive alongside heightened safety work in the field, including Google DeepMind's AI Control Roadmap, which proceeds from the premise that AI should not be assumed to always act as intended.
Category highlights
Foundation Models: GLM-5.2, Step 3.7 Flash, Gemma 4
Zai.org's GLM-5.2 (1M context, stronger agentic coding) launched on Together AI, while StepFun's Step 3.7 Flash is free for a month in Cline and MiniMax M3 spread across inference platforms. Google's Gemma 4 31B demonstrated autonomous database optimization as an ADK agent, and Cohere published three updates improving accessibility of North Mini Code.
Audio: Grok TTS Leads Humanness Index, Sony AI Ships Woosh
Grok TTS topped Vapi's Humanness Index, and Sony AI released Woosh, a sound-effects foundation model for professional libraries targeting games and film. Async Pro v1.0 began offering six-language multilingual TTS via fal, and Pixel Film Studios launched an AI audio mastering plugin for Final Cut Pro.
Platforms: Grok on Databricks, Perplexity 'Brain in Computer'
Grok was integrated into Databricks Agent Bricks, and Perplexity announced a continual-learning memory called 'Brain in Computer.' fal expanded serverless analytics features, and the 10B-class open image model Boogu Image debuted.
Agents Move Toward Reusable Skills and Long-Horizon Autonomy
OpenAI Codex's 'Record & Replay' converts task demos into editable skills, Kimi's 'Goal Mode' runs around the clock, and Luma Skills and Pika MCP skills follow the same pattern of turning workflows into reusable skills. Luma Skills lets users upload creative DNA once for consistent, repeatable generation and team sharing.
Databricks Data+AI Summit 2026 and NeurIPS Ethics Reviewers
The Databricks Data+AI Summit 2026 took place, with Stanford winning the Grounded Reasoning Cup. NVIDIA shared its French AI factory deployment at VivaTech, and NeurIPS 2026 is recruiting Ethics Reviewers.
Key trends
AI Power Demand Emerges as Industry and Policy Constraint
Growing data center demand and national energy policies, including a return to nuclear power, are surfacing as constraints on AI expansion and points of industry and policy debate.