Skip to content
Daily Brief News Daily Brief

9 AI stories from June 19, 2026: Cursor v3.8 automate skill, DeepMind AI Control Roadmap, Noam Shazeer joins OpenAI, ChatGPT health intelligence upgrade, Grok on Databricks, Amazon Trainium3 external sales, Adobe creative agents, FERC grid fast-track, and Mistral sovereignty pitch

· by Pondero Newsdesk · 9 stories

AI news daily brief: 2026-06-19

Nine stories today: Cursor ships v3.8 with a new automation skill and six GitHub triggers, Google DeepMind publishes a defense-in-depth agent security framework, Noam Shazeer moves from Google to OpenAI ahead of the IPO, ChatGPT gets a physician-backed health intelligence upgrade, Grok lands on Databricks Agent Bricks, Amazon signals plans to sell Trainium3 chips externally, Adobe rolls out creative agents across Creative Cloud, FERC orders fast-track grid connections for data centers, and Mistral's CEO frames the Anthropic export ban as a case for European AI sovereignty.

Cursor v3.8 ships /automate skill, Slack emoji trigger, and five new GitHub triggers for cloud automation sessions

Cursor released version 3.8 on June 18 with a new /automate skill that accepts a plain-language goal and configures the triggers, instructions, and tools automatically. Six new trigger types ship alongside it: a Slack emoji trigger, plus five GitHub triggers for issue comments, PR review comments, PR review submissions, review thread updates, and workflow completion. Cloud agents in Automations can now use their own computers to produce demos and artifacts, per the Cursor v3.8 changelog.

Full story: Cursor v3.8 ships /automate skill and six new automation triggers

Google DeepMind publishes AI Control Roadmap treating internal agents as insider threats

Google DeepMind published its AI Control Roadmap on June 18, a defense-in-depth framework that plans for a capable, misaligned agent actively trying to evade oversight. The framework uses the MITRE ATT&CK taxonomy to classify AI threats and deploys "trusted AI systems" as real-time supervisors that can block harmful agent actions before execution. Detection levels run D1 to D4; response levels run R1 to R3. The roadmap notes its own limits: "This monitoring only works as long as the monitored models don't learn to game it." DeepMind released it as a candidate industry standard, per the DeepMind security blog post.

Full story: Google DeepMind AI Control Roadmap

Noam Shazeer leaves Google to join OpenAI as Lead for Architecture Research ahead of IPO

Noam Shazeer announced on June 18 that he is leaving Google to join OpenAI as Lead for Architecture Research. Shazeer co-authored the 2017 "Attention Is All You Need" paper that introduced the Transformer. He left Google in 2021 to co-found Character.AI, returned in 2024 as Gemini VP of Engineering when Google licensed Character.AI technology for approximately $2.7 billion, and will now focus on next-generation model architectures at OpenAI. Dean Ball, a former Trump administration AI policy official, also announced he will join OpenAI on July 6 to lead a new "Strategic Futures" team, per TechCrunch.

Full story: Noam Shazeer joins OpenAI

OpenAI upgrades ChatGPT health intelligence with a 262-physician Global Physician Network and HealthBench

OpenAI announced on June 18 that it has upgraded ChatGPT's health capabilities using feedback from a Global Physician Network of 262 doctors spanning 26 specialties and 59 countries. The physicians scored more than 600,000 model responses across accuracy, safety, communication clarity, context awareness, completeness, and escalation judgment. The resulting improvements to GPT-5.5 Instant produced what OpenAI describes as a 71% reduction in flagged health factuality issues, per its announcement page. OpenAI also introduced HealthBench, a physician-authored evaluation framework intended as a repeatable test for health intelligence claims. Per OpenAI, more than 230 million people use ChatGPT for health-related questions each week.

Full story: OpenAI ChatGPT health intelligence upgrade

xAI Grok goes live on Databricks Agent Bricks at the 2026 Data + AI Summit

xAI and Databricks announced on June 18, at the final day of the Databricks Data + AI Summit in San Francisco, that Grok models are now natively available on Databricks Agent Bricks. Enterprise teams running data pipelines on the Databricks Lakehouse can use Grok directly against their structured and unstructured data with zero data retention for model training. Per the xAI announcement, Agent Bricks processes 1 quadrillion tokens annually for enterprise customers including AstraZeneca, 7-Eleven, Fox Corporation, and Block. The summit also featured the OpenSharing protocol launch, a new Linux Foundation open standard for sharing AI agent skills, models, and unstructured data across platforms, backed by OpenAI, SAP, Stripe, Atlassian, and others.

Full story: xAI Grok on Databricks Agent Bricks

Amazon in talks to sell Trainium3 chips to outside data centers in first-ever external hardware sales

AWS AI chief Peter DeSantis told Bloomberg on June 18 that Amazon is in active talks to sell its Trainium3 AI chips directly to other companies for use in their own data centers, a shift from cloud-only deployment. Trainium3 delivers four times the performance of Trainium2 at roughly half the cost of comparable GPUs and has run near capacity since its late-2025 launch, per TechCrunch. Amazon estimates the external chip market represents a $50 billion annual revenue opportunity. If the talks result in sales, it would be the first time Amazon sells proprietary AI accelerator hardware outside of AWS, directly competing with Nvidia's hardware business model. OpenAI and Anthropic have already committed $225 billion in Trainium compute contracts.

Adobe ships AI creative agents into Photoshop, Premiere, Illustrator, and InDesign with Claude, ChatGPT, and Copilot integrations

Adobe rolled out creative agents across Creative Cloud on June 18, enabling multi-step production workflows through natural-language descriptions. The agents handle tasks in Photoshop (background swaps, platform resizing), Premiere (footage organization, rough cuts), Illustrator (versioned file generation from spreadsheets), InDesign (brand-guideline layout updates), and Frame.io (footage consolidation). Adobe simultaneously launched Claude, ChatGPT, and Microsoft 365 Copilot integrations for the agent system, with Google Gemini and Slack integrations announced as coming soon, per The Decoder.

FERC orders six major grid operators to fast-track AI data center connections with 30- and 60-day deadlines

The Federal Energy Regulatory Commission issued orders on June 18 requiring the six major US grid operators, including PJM, the country's largest, to fast-track interconnection requests from data centers and other large electricity consumers. Grid operators have 30 days to report available generating capacity and 60 days to defend or revise regional electricity rates to accommodate large load additions. The orders also direct operators to consider alternative transmission technologies and accommodate behind-the-meter power solutions, per TechCrunch. The commission did not address the underlying generating capacity shortage, so the orders accelerate the approval process without resolving the multi-year interconnection queues at most large operators.

Mistral CEO pitches European AI sovereignty after Anthropic export ban and eyes 20 billion euro valuation

Mistral CEO Arthur Mensch responded publicly on June 17 to the Trump administration's export control order against Anthropic's Fable 5 and Mythos 5 models by positioning Mistral as the sovereign alternative for European enterprises. "We exist outside of state control," Mensch said, arguing the episode illustrates the risk of European governments depending on US-controlled AI subject to unilateral government directives. Separately, Mistral is in early talks to raise approximately 3 billion euros at a valuation of roughly 20 billion euros, per Bloomberg reporting from June 12, which would nearly double its September 2025 Series C valuation. Confirmation of a lead investor and any round close is the next development to track.

Sources