AI News: June 26, 2026

1. White House Asks OpenAI to Stagger GPT-5.6 Release Over Cyber Concerns

The White House. The Trump administration asked OpenAI to roll out its forthcoming GPT-5.6 model gradually rather than launching it publicly all at once, citing safety and security concerns raised by the Office of the National Cyber Director and the Office of Science and Technology Policy. Officials worried that frontier cyber-capable models could be exploited to identify software vulnerabilities or accelerate ransomware attacks faster than humans can respond. Under the plan, GPT-5.6 would first reach a vetted set of partners approved customer by customer, with a broader public release roughly two weeks later if the preview goes well. Source

2. Qualcomm Enters the Data Center Market With Dragonfly C1000 and $4B Modular Buy

Qualcomm. The company unveiled the Dragonfly C1000, a data center processor optimized for AI agents that emphasizes high performance at low power, with Meta committing to deploy it starting in 2028. Qualcomm also announced a roughly $4 billion acquisition of AI software startup Modular, whose toolchain lets AI applications run across different chip architectures, complementing its hardware push. Qualcomm nearly doubled its non-smartphone revenue forecast to $40 billion by 2029 (including $15 billion from data centers), and its stock jumped 15 percent in after-hours trading. Source

3. General Intuition Raises $320M to Train Agents on Video Game Footage

General Intuition. The startup raised $320 million at a $2.3 billion valuation in a round led by Khosla Ventures, with backers including General Catalyst, Jeff Bezos, Eric Schmidt, and researchers from Google DeepMind and MIT. It trains AI agents on hundreds of millions of hours of gameplay from parent company Medal to teach spatial-temporal reasoning, aiming for a single model that handles both in-game situations and real-world dynamics for robots and embodied systems. The bet is that game environments can serve as scalable training grounds for agents that transfer to the physical world. Source

4. Patronus AI Raises $50M to Build Simulated Worlds That Stress-Test Agents

Patronus AI. The agent-evaluation startup closed a $50 million Series B led by Greenfield Partners, with Notable Capital, Lightspeed, Datadog, and Samsung participating, bringing total funding to $70 million. Patronus builds replicas of websites and internal systems where AI agents are stress-tested via reinforcement learning, currently focused on software engineering and finance use cases. Founded in 2023 by former Meta AI researchers, the company says it serves nearly every frontier AI lab as customers. Source

5. Databricks’ Former AI Chief Bets on a New Chip to Cut Inference Power 1,000x

Unconventional AI. Naveen Rao, the former head of AI at Databricks, launched Unconventional AI to attack what he calls AI’s fundamental energy limit, claiming an oscillator-based computer architecture can cut inference power consumption by 1,000x versus conventional chips. The company’s first product, an image generator called Un-0, reportedly matches state-of-the-art diffusion models like Stable Diffusion in software simulation of the new architecture. Rao framed the effort as the “hello world of a new kind of computer,” arguing energy will be the binding constraint on AI scaling in the coming years. Source

6. Adobe Acquires AI Enhancement Tool Maker Topaz Labs

Adobe. The company agreed to acquire Topaz Labs, a two-decade-old maker of AI-powered image and video enhancement tools, with the deal expected to close in the second half of 2026 and no price disclosed. Topaz brings proprietary models including Astra for video upscaling and Wonder for image retouching, plus technology that runs large AI video models on consumer-grade GPUs. Adobe highlighted Topaz’s expertise in optimizing complex AI models to run directly on device, pointing to faster, more accessible tools for creative professionals. Source

7. Amazon Commits a Fresh $13B to AI Infrastructure in India

Amazon. The company announced an additional $13 billion commitment through 2030 to expand AWS data center capacity in Mumbai and Hyderabad, following CEO Andy Jassy’s meeting with Indian Prime Minister Narendra Modi. The pledge brings Amazon’s total planned India investment to $48 billion and lands amid a broader race by Microsoft and Google to build AI infrastructure hubs in the country. The scale underscores how India is becoming a key battleground for global cloud and AI capacity buildout. Source

8. Washington Post Study Finds Most AI Chatbots Still Lean Left

The Washington Post. An investigation testing six leading models on political questions found persistent leftward leanings across most platforms, with OpenAI’s GPT-5.5 giving exclusively left-leaning answers 80 percent of the time and Deepseek V4 Pro 70 percent. Anthropic’s Claude Opus 4.8 leaned left 43 percent of the time while presenting both sides in 57 percent of cases, and even right-positioned models like xAI’s Grok 4.3 and Gab’s Arya skewed left more often than expected. Google’s Gemini 3.1 Pro stood out as the exception, presenting both perspectives in 93 percent of cases. Source

9. Authors Guild Test Finds AI Detectors Wildly Inconsistent on Human Writing

The Authors Guild. The writers’ organization tested five AI detection tools against ten human-written articles from 2020 to 2022 and found drastically different reliability, with Pangram and Grammarly correctly flagging every piece as human and zero false positives, while Sidekicker labeled all of them AI-generated. ZeroGPT produced inconsistent, unreliable scores, reinforcing the Guild’s warning that detectors should never solely determine publishing decisions because false positives can damage author reputations and contracts. The test measured only false positives on human text and did not confirm the tools reliably catch AI-generated writing. Source