AI News Summary — Week 25.1, 2026

This week in AI: Major talent shakeups as Google loses two flagship researchers, the US government bans Anthropic's most capable models, and Europe responds with calls for independent AI infrastructure.

Google Loses Two Top AI Researchers to OpenAI & Anthropic

Google DeepMind just got a double dose of bad news. Within days of each other, two of its most prominent researchers announced they're leaving for rival labs — and the market responded harshly.

Noam Shazeer, co-lead of the Gemini models and co-author of the seminal "Attention Is All You Need" paper that introduced the Transformer architecture powering today's LLMs, is heading to OpenAI. He was brought back to Google in 2024 through a deal with Character.AI reportedly valued at $2.7 billion. Less than two years later, he's leaving.

John Jumper, who led the AlphaFold project and won the 2024 Nobel Prize in Chemistry for his work on protein structure prediction, is moving to Anthropic. He spent nearly nine years at DeepMind. His move aligns with Anthropic's growing focus on AI for science — they have an AI for Science event scheduled for June 30.

Alphabet shares fell about 5-6% on June 22, with market reports tying the drop to concerns about AI spending and Google's ability to retain senior AI talent.

Why It Matters

Bloomberg reported that staff at DeepMind have raised concerns about the company lacking a clear product for businesses building AI coding tools — an area where Anthropic and OpenAI have gained ground. This echoes what CEO Sundar Pichai said in May: Google was "a bit behind" on agentic coding.

Where researchers of this caliber choose to work shapes how investors and competitors read the frontier AI race. Google paid a premium to bring Shazeer back once and couldn't keep him. That's a data point.

US Government Bans Anthropic's Claude Models

In a dramatic escalation, the federal government told Anthropic on June 12 that its Claude Fable 5 and Mythos 5 models pose a threat to national security and placed export controls on them. Anthropic revoked access to both models hours later.

The backstory: In April, Anthropic announced it had built Mythos — a model so capable at cybersecurity that it could pose a global threat. They gave access to a small group of verified cybersecurity experts. Then they released Fable 5 (the guarded version) on June 9. Four days later, the ban came.

The Backlash

The move has sparked controversy on multiple fronts:

Cybersecurity experts are speaking out. An open letter signed by leading researchers argues that blocking access to these models actually makes the US more vulnerable, not less. They say the models were helping defenders prepare for attacks, and that Fable 5 is no more dangerous than other leading models already widely available.
Europe sees a wake-up call. French politician Bruno Retailleau described the ban as a signal that Europe needs its own AI capabilities. The irony: China's open-source models are cheap, powerful, and come with no US-imposed restrictions. Chinese startup Zhipu has seen shares skyrocketing.
Amazon may have triggered this. Reports indicate Amazon CEO Andy Jassy was the one who told government officials that Fable would be dangerous. Amazon is both invested in Anthropic and building its own competing AI models.

The Bigger Picture

This is the second feud between Anthropic and the government this year. Following the last clash over Pentagon usage, a slate of new bills was introduced defining limits on military AI. With each drastic action, pressure for federal AI regulation grows. Polling shows most Americans want it.

The question now: What happens if companies decide working with Chinese models is just easier? Could the next drastic decision be that US companies using Chinese models pose a national security threat?

Google Builds Trusted Directory for AI Agents

In other Google news, the company is developing a Trusted Directory for AI agents, according to reports this week. The initiative appears aimed at establishing trust and verification standards for AI agents operating across platforms — an attempt to bring order to an increasingly agentic web.

What's Hot This Week

Researcher exodus: Noam Shazeer to OpenAI, John Jumper to Anthropic — Alphabet stock drops 5-6%
Anthropic ban: US government places export controls on Claude Fable 5 and Mythos 5
Open letter: Cybersecurity experts say the ban makes the US more vulnerable
Europe's response: Politicians call for independent AI infrastructure
Agent standards: Google building a trusted directory for AI agents

That's a wrap for Week 25.1. Stay tuned!

Sources: MIT Technology Review, Search Engine Journal, TechCrunch, CNBC, Bloomberg