Google DeepMind's Noam Shazeer Joins OpenAI, Anthropic's Fable 5 Restored With New Controls
A major talent shift sees a Transformer co-author move to OpenAI, while Anthropic's flagship model returns online under new government-mandated restrictions, reshaping access to frontier AI.
The story
The AI software landscape saw a significant talent migration this week as Noam Shazeer, a co-author of the foundational Transformer paper and co-lead for Google DeepMind's Gemini models, announced his departure to join OpenAI as Lead for Architecture Research. This move comes after Google reportedly paid $2.7 billion in 2024 to bring Shazeer back from Character.AI, which he co-founded.
Meanwhile, Anthropic's Claude Fable 5 model was restored on Thursday, June 18, after a six-day government-forced shutdown. The model now includes nationality-based access controls and tightened safety classifiers, with its unrestricted cybersecurity variant, Mythos 5, remaining under separate, stricter government-supervised access. These developments highlight both the intense competition for top AI talent and the growing regulatory scrutiny impacting the deployment and accessibility of advanced AI systems.
Who moved
OpenAI
What Changed: Noam Shazeer, co-author of the Transformer paper and co-lead of Google DeepMind's Gemini, joined OpenAI as Lead for Architecture Research.
Consequence: This marks a significant gain in top-tier talent for OpenAI, bringing deep expertise in foundational AI architectures.
Anthropic
What Changed: Claude Fable 5 was restored with new nationality-based access controls and tightened safety classifiers after a six-day government-forced shutdown.
Consequence: The restoration, though with new restrictions, re-enables access to a powerful model for enterprise customers, albeit with increased compliance screening.
Google DeepMind
What Changed: Lost Noam Shazeer, a key architect of its Gemini models, to OpenAI.
Consequence: This departure creates a vulnerability for the upcoming Gemini 3.5 Pro launch, given Shazeer's deep understanding of the model's architectural decisions.
SpaceX
What Changed: Acquired AI coding assistant company Anysphere, makers of Cursor, in an all-stock transaction valued at $60 billion.
Consequence: This represents the largest acquisition of a venture-backed startup in history, significantly expanding SpaceX's AI software capabilities, particularly in coding.
Databricks
What Changed: Debuted a new suite of AI agents called Genie One at its Data + AI conference.
Consequence: These agents are positioned as 'agentic coworkers' and aim to become the next-generation system of record for enterprises, leveraging Databricks' data platform.
Morocco
What Changed: Launched the Rally AI Future Lab, a national initiative to foster homegrown AI talent.
Consequence: The program aims to develop domestic AI capability, prototype solutions, and create a pipeline from idea to operational product, with an ambition to reach 5,000 participants.
New models
Gemini 2.5 Flash
Lab: Google
What: Became the default model across all Gemini products.
Use: This makes Google's faster, lighter Gemini model more widely accessible for general use cases.
Genie One
Lab: Databricks
What: A new suite of AI agents designed for enterprise use.
Use: Enterprises can use these agents as 'agentic coworkers' to automate tasks and build next-generation systems of record.
Market signals
DeepSeek reportedly raised $7.4 billion in funding, pushing its valuation to $50 billion.
Implication: This massive funding round underscores continued strong investor appetite for frontier AI companies.
Baseten reportedly secured a $1.5 billion split-priced funding round.
Implication: This signals continued investment interest in AI inference platforms that optimize speed and cost control for AI models.
ChatGPT's market share dipped below 50% for the first time since its launch.
Implication: Users are increasingly migrating between different AI assistants, indicating a diversifying and competitive chatbot market.
China announced a $295 billion, five-year national AI infrastructure investment plan.
Implication: The plan, requiring 80% domestic technology, indicates a strategic push for technological sovereignty and self-reliance in AI development.
Aether AI raised $20 million in seed funding for causal AI.
Implication: This investment highlights a growing belief that the next frontier in AI for applications like robotics requires models understanding cause-and-effect, not just correlations.
Builders saying
Yann LeCun, founder of AMI Labs and former chief AI scientist at Meta. LeCun called xAI a 'failure' and warned of an AI bubble, citing its economics and difficulty in hiring top talent.
Andrej Karpathy, Anthropic's head of pretraining research. Karpathy was personally locked out of Anthropic's Fable 5 and Mythos 5 models due to the US government's export control directive, highlighting the abrupt impact of regulatory actions on researchers.
Dario Amodei, CEO of Anthropic, and Demis Hassabis, CEO of Google DeepMind. They called for a US-led AI coalition at the G7 summit to coordinate AI governance across borders and establish global standards.
What we'll be watching
- Google is expected to release Gemini 3.5 Pro, featuring a 2 million token context window, before June 30.
- Enforcement obligations for chatbot systems under the EU AI Act take effect on August 2, 2026.
- AMD AI DevDay 2026 will bring together engineers and developers for technical sessions and workshops on AI workloads.
- The AI Conference 2026 will take place in San Francisco from September 29 to October 1, gathering builders, researchers, and leaders.
Reporting + analyst voices: grounded via Google Search at publish time.