OpenAI Is Bringing On Some Big Guns in the Lead-Up to Its IPO — TechCrunch Noam Shazeer — co-author of the landmark 2017 “Attention Is All You Need” paper and VP of Engineering / Gemini co-lead at Google DeepMind — is leaving to join OpenAI as Lead for AI Architecture Research. Sam Altman called it “10 years in the making.” The move comes less than two years after Google reportedly spent $2.7 billion to bring Shazeer back from Character.AI, marking a stinging talent loss for Alphabet and a coup for OpenAI ahead of its IPO.

🚨 Policy & Safety

Anthropic Employees Accuse Trump Administration of Targeting Them — The New York Times

More than 150 cybersecurity experts have signed an open letter calling on the Trump administration to lift export restrictions on Anthropic’s Fable 5 and Mythos models. Employees argue the ban — predicated on a narrow claimed jailbreak — ignores extensive protections built into Fable specifically to prevent offensive cyber use. The dispute traces back to Anthropic’s refusal to allow the U.S. military to deploy its models for fully autonomous weapons systems, after which the Pentagon reportedly blacklisted the company.

Who Decides When AI Is Too Dangerous? — The Verge (Decoder) The Verge’s Decoder digs into the Fable 5 saga with senior AI reporter Hayden Field, tracing how a “fix this code” prompt became the basis for U.S. export controls and what it signals for AI governance going forward.

Amazon Employees Say They’re Facing Termination for Backing Data Center Limits — The Verge Three Amazon engineers who testified at Seattle City Council hearings about a proposed data center moratorium now face disciplinary action — one week after their testimony — raising questions about retaliation for protected political speech.

AI Data Centers Just Got a Government-Mandated Fast Lane to the Grid — TechCrunch

🤖 Frontier Models

ChatGPT’s Market Share Slips Below 50% for the First Time — TechCrunch

ChatGPT has lost its majority hold on the AI assistant market for the first time, with users spreading across Gemini, Claude, and Grok. Despite being the fastest app ever to reach a billion monthly users, the shift signals users are increasingly comfortable switching — and that competitors have closed the gap in quality perception.

Cursor’s New Model — Cursor / Anysphere Cursor revealed a 1.5-trillion-parameter coding model trained from scratch on more than 100,000 GPUs, designed to push agentic software development well beyond autocomplete. A public release is expected in the coming weeks.

Mistral Has a New Model Coming This Summer — Mistral

Kimi K2.7 Code vs Claude Fable 5: Landing Pages That Cost 94% Less — @nutlope

🛠️ Developer Tools

Anthropic Ships Major Claude Design Overhaul — VentureBeat

Claude Design has been rebuilt into an enterprise platform with support for design system imports from GitHub, design files, or raw uploads; bidirectional code round-trips; and a fix for its token-burning behavior. The update transforms it from a prototype toy into a tool for generating consistent branded output at scale.

Vercel Launches eve and Connect for AI Agent Infrastructure — Vercel

Vercel shipped two complementary agent products today. Eve is an open-source agent framework with built-in durable execution, sandboxed compute, approvals, and subagents. Connect, now in public beta, replaces long-lived provider tokens with runtime credential exchange that issues short-lived, task-scoped credentials — reducing exposure from shared persistent access tokens.

Claude Code: Artifacts, Steering Guide, and Enterprise MCP Auth — Anthropic Anthropic published three Claude Code updates today: artifacts support, a comprehensive guide to steering Claude Code via CLAUDE.md / skills / hooks / subagents, and centralized enterprise MCP authorization management.

Claude Code v2.1.181 — Anthropic

Replit Is Now Available in Claude — Replit

🎨 Creative Tools

Photoshop and Premiere Now Have AI Assistants — The Verge

Adobe launched bespoke AI assistants across its major Creative Cloud apps in a public beta today — Photoshop, Premiere, Illustrator, InDesign, and Frame.io each get their own workflow-tuned AI chatbot. Alongside this, Adobe launched a redesigned Firefly AI studio in private beta, with persistent project context, reusable assets, and a unified generate-and-edit interface.

Midjourney Goes from Generating Cat Images to Full-Body Ultrasound Scans — The Verge Midjourney CEO David Holz unveiled The Midjourney Scanner, a hardware full-body ultrasound device using a ring of sensors — a sharp pivot from the AI image generation the company is known for.

🔬 Research

Introducing LifeSciBench — OpenAI OpenAI released LifeSciBench, a benchmark evaluating AI on end-to-end life sciences workflows — evidence analysis, experimental design, scientific reasoning, and research communication — rather than isolated biology Q&A tasks.

Are LLMs Thonking Too Hard? — parsiya.net A follow-up security study testing 26 combinations of Claude 4.6/4.7 and GPT 5.4/5.5 finds that higher reasoning effort and newer model versions are not reliably better at triaging security vulnerabilities — a counterintuitive result for teams investing in reasoning-heavy pipelines.

MolmoMotion: Language-Guided 3D Motion Forecasting — Allen AI / Hugging Face

Generated by claude-sonnet-4-6 · 2026-06-18T10:00:00Z