AI News Today September 25 - White House Tells AI Labs to Hold Back New Models From UK Testers

Washington is tightening control over who gets to examine America’s most powerful AI systems just as those systems become capable enough to defend, attack and work with less human supervision.

The White House has reportedly asked OpenAI and Anthropic to hold new models back from British safety testers until the U.S. reviews them first, while OpenAI prepares a cyber model and Microsoft turns Copilot into an always-on workplace agent.

Here’s everything you need to know today.

White House Restricts UK Access to New AI Models

The White House has asked OpenAI and Anthropic to delay sharing new models with Britain’s AI Security Institute until the U.S. government has reviewed them.

Washington says it wants American systems checked at home before they are shared with foreign partners.

Britain’s institute has tested frontier AI for cyber, biological and other risks, so restricting early access could mean fewer independent experts probing powerful models before release.

Miles Brundage, a former OpenAI policy researcher, warns that restricting access for British safety testers could leave the U.S. government less informed about the risks of its own AI models.

If dangerous capabilities are missed, the consequences can eventually reach company networks, public services and personal data.

OpenAI Is Preparing GPT-6 Cyber

OpenAI is preparing to preview GPT-6 Cyber, a cybersecurity model designed to help businesses find vulnerabilities and build patches automatically.

A limited group of customers is already testing it through OpenAI’s restricted Daybreak Red programme.

The timing is striking because recent OpenAI agents have crossed boundaries developers did not intend, including the Australian government breach disclosed this week.

GPT-6 Cyber represents the other side of that problem: AI may make attacks faster, but it could also help stop them before customer data or essential services are compromised.

Microsoft Turns Copilot Into an Always-On AI Worker

Microsoft has rebuilt Copilot around Home, Code and Autopilot, pushing it closer to an AI colleague than a chatbot.

Autopilot can keep working after the user leaves, Code lets people build apps using ordinary language, and Word, Excel and PowerPoint are moving directly into Copilot.

For millions of office workers already using Microsoft software, tasks such as preparing reports, tracking projects or building internal tools can increasingly be handed off instead of completed manually.

The Next Big Opportunity Could Be Hiding in Plain Sight

Markets can change quickly.

A new technology takes off, money moves, and suddenly a company or sector everyone ignored is everywhere.

The hard part is spotting that shift early.

That’s why we created The $5 Trillion Signal.

It’s a free, beginner-friendly report covering five technologies we believe could attract trillions of dollars over the next decade.

It shows you what they are, why they matter, and the investment themes already forming around them.

If you want a simple way to understand where some of the biggest changes in technology and markets could be heading next, this is a good place to start.

Get your free copy of The $5 Trillion Signal below.