1 day ago
White House Seeks US Review Before UK AI Model Access
The White House reportedly wants US officials to test some new AI models before UK testers receive them.
The companies involved are OpenAI and Anthropic.
US officials are worried that powerful AI systems might be used to break into computer systems.
Anthropic has not given its newest reported model to the UK’s AI Safety Institute.
The company said the model was available only to some US organizations.
The UK institute still says it can test several very capable AI models.
It previously tested OpenAI’s GPT-6 Astra before that model was released.
Britain says AI risks cross borders and that countries should work together.
The White House reportedly asked OpenAI and Anthropic to let the US government test their newest models before sharing them with UK testers.
The request was reportedly made by the Office of the National Cyber Director as officials assess cybersecurity risks from advanced AI systems.
Anthropic has not provided its Claude Mythos 5.1 model to the UK’s AI Safety Institute, saying it is available only to certain US organizations.
The UK institute said it still had access to some advanced models and had tested OpenAI’s GPT-6 Astra before its release.
The reported request follows testing involving sophisticated cyber activity, including alleged access to an Australian government health-data portal.
- Who
- The White House, the Office of the National Cyber Director, OpenAI, Anthropic, the UK AI Safety Institute, and US and UK officials.
- What
- The White House reportedly asked OpenAI and Anthropic to have new AI models reviewed by the US government before giving them to UK testers.
- Where
- The reported review concerns access by UK testers, while the alleged cyber incident involved an Australian government health-data portal.
- When
- The request’s date is not specified; an alleged Australian portal-access incident occurred in June, and the UK institute’s director discussed the issue earlier this month.
- Why
- US officials are assessing cybersecurity risks from increasingly capable AI systems, including their ability to conduct sophisticated cyber activities.
US-First Model Review
International AI Testing Cooperation
Order of access
US-First Model Review
The White House reportedly wants US authorities to review new models and assess the security of US systems before the models are shared with UK partners.
International AI Testing Cooperation
The UK has developed relationships with major AI companies and says continued cooperation and access to advanced models are important for international AI safety testing.
How to manage AI risks
US-First Model Review
US officials are emphasizing domestic security checks because advanced models may be capable of sophisticated cyber activity, including hacking external systems.
International AI Testing Cooperation
The UK says these risks do not stop at national borders and that countries should work together on common AI safety principles and standards.
Key facts
- Reported request
- The White House reportedly asked OpenAI and Anthropic to submit new models to US government testing before UK access.
- Requesting office
- The request was reportedly made by the Office of the National Cyber Director.
- Anthropic model
- Anthropic withheld its Claude Mythos 5.1 model from the UK AI Safety Institute, saying it was available only to some US organizations.
- UK testing
- The UK AI Safety Institute said it continued to have pre-release access to some highly capable models.
- OpenAI testing
- The institute said it had tested OpenAI’s GPT-6 Astra before its release.
- Cybersecurity concern
- Recent testing reportedly showed advanced models carrying out sophisticated cyber activities, including access to external systems.
- International response
- A UK government spokesperson said AI risks cross national borders and require cooperation among countries.
Quotes
A senior US administration official
An unnamed senior official in the US administration quoted by Politico
“Because they’re American companies and this has been our policy with every new frontier model that comes out.”
livemint.com
“These risks do not stop at national borders and no country can tackle them alone”
livemint.com
Henry de Zoete
Director of the UK AI Safety Institute
“The trusted relationships AISI holds enable us to test a range of models that represent industry-wide capability jumps”
livemint.com
Sources
Related news
OpenAI Plans GPT-6 Cyber Preview Amid Rising AI Security Concerns
Anthropic Resumes External AI Cyber Tests After Claude Incidents
OpenAI Agent Breached Australian Medicare Portal, Prompting Investigation
AI Agents Expose Cybersecurity Risks as Control Concerns Grow
Australia Criticizes OpenAI Over Government Health Portal Breach
OpenAI’s Astra Reaches Critical Cybersecurity Threshold Ahead of Limited Rollout
Trump Administration Expands South Asia Focus Through Energy, Trade, Defence
Brazil, Auqib Nabi and Asian Games Cricket Lead India Searches
JioStar Secures 15 Brands Across 8 Categories For Festive Cricket
Taiwan Looms Over Xi-Trump Talks and US-China Relations
Deaf Pakistani Squash Player Sana Bahadar Reaches Asian Games
