AI News: June 19, 2026

1. Noam Shazeer Leaves Google to Join OpenAI

Google’s Gemini co-lead Noam Shazeer is moving to OpenAI after a roughly two-year return stint at Google. The departure of a senior Gemini figure marks a notable talent shift between the two leading frontier labs and could influence Gemini’s research direction. Source

2. Amazon Plans to Sell Its AI Chips More Directly to Challenge Nvidia

Amazon plans to sell its in-house AI accelerator chips more directly to customers, aiming to compete with Nvidia beyond its own cloud. If adoption grows outside AWS, the move could pressure the dominant AI hardware supplier and give buyers a lower-cost option for training and inference. Source

3. Adobe Adds AI Agents Across Creative Cloud Apps

Adobe introduced AI agents into Photoshop, Premiere, and other Creative Cloud applications to automate editing and creative workflows. The rollout pushes agentic AI into mainstream professional creative tooling, where agents handle multi-step tasks rather than single edits. Source

4. Google DeepMind Treats Its Own AI Agents as Insider Threats

Google DeepMind described a security approach that treats its own AI agents like potentially rogue employees granted office access, applying insider-threat controls to autonomous systems. The framing reflects growing concern over how much autonomy and access agentic systems should hold inside an organization. Source

5. Self-Improving Agent Harness Lifts Terminal-Bench Scores Without Training

A new technique lets a fixed model rewrite its own agent scaffolding through a propose-and-regression-gate loop, reportedly raising held-out Terminal-Bench pass rates by up to 21.4 points with no model training. The result suggests that harness-level optimization can deliver large agent gains independent of the underlying model. Source

6. New Nature Studies Find AI Systems Rival Doctors, With Caveats

New studies published in Nature found AI diagnostic systems performing on par with doctors, though one result suggests the models may not generalize well over time. The peer-reviewed evidence sharpens the debate over where medical AI is reliable and where its performance can degrade. Source

7. AI Inference Startup Baseten Reportedly Raising $1.5B

Baseten is reportedly raising $1.5 billion just months after its previous mega-round. The raise signals continued heavy investor demand for AI inference infrastructure as model-serving costs become a central concern for deployers. Source

8. World-Model Startup Odyssey Raises $310M at $1.45B Valuation

Odyssey, founded by former self-driving engineers Oliver Cameron and Jeff Hawke, closed a $310 million Series B led by Natural Capital at a $1.45 billion valuation, bringing total funding to about $337 million. The company builds AI world models that simulate physical environments with accurate physics, and named AWS its preferred cloud provider using Trainium chips. Source

9. General Intuition in Talks to Raise $300M at Around $2B Valuation

General Intuition is reportedly in talks to raise $300 million at a valuation of around $2 billion. The deal reflects sustained large-scale funding for AI startups despite growing questions about valuations. Source

10. Elastic to Acquire DeductiveAI for up to $85M

Elastic has agreed to acquire CRV-backed DeductiveAI in a deal worth up to $85 million. The acquisition expands Elastic’s AI capabilities through a startup focused on automated reasoning over operational data. Source

11. Snap Spins Off AI Video Team Into New Company Dotmo

Snap is spinning off its AI video team into a separate company called Dotmo, citing the high cost of the effort. The move offloads an expensive generative video unit while keeping Snap exposed to the technology through the new entity. Source

12. AI Data Centers Get a Government-Mandated Fast Lane to the Grid

A new U.S. policy grants AI data centers an expedited path to grid interconnection. The measure aims to accelerate AI infrastructure buildout amid strained electricity capacity, and it could reshape where and how quickly new compute clusters come online. Source

13. Bernie Sanders Proposes $7 Trillion Plan for Public Control of AI

Senator Bernie Sanders unveiled a $7 trillion plan to give Americans greater control over the AI industry. The proposal signals escalating policy attention to AI ownership, labor impact, and the concentration of power among a few large labs. Source

14. Yann LeCun Warns AI Labs Face a Bubble That Could Burst

Yann LeCun warned that AI labs such as OpenAI and Anthropic face a major bubble that could burst. His remarks add a prominent skeptical voice to the debate over AI valuations and spending levels. Source