Reuters: US withdraws request over British AI testers
British safety researchers may again gain access to US models before their public launch.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
_-1.jpg){kind=link}
The Trump administration has withdrawn an earlier request not to allow US AI companies to share new models with British safety researchers, according to Reuters. The change of course could restore cooperation between the UK AI Security Institute and US companies, but has not been officially confirmed by all parties involved.
US AI developers may again give the British government early access to new models for safety testing, according to Reuters. Reuters bases this on reporting by The Telegraph and writes that the US government has reversed an earlier request. That request is said to have asked companies such as OpenAI and Anthropic not to share new models with the AI Security Institute before their public launch.
The AI Security Institute, part of the Department for Science, Innovation and Technology, tests advanced models for capabilities and risks. The organisation says it examines models before their public introduction and works with AI companies in doing so. Such tests look, among other things, at cyber capabilities, deception, autonomy and ways in which safety measures can be circumvented.
According to Reuters, the earlier US restriction was linked to Washington’s wish to oversee the sharing of new models itself first. The new policy does not mean that a formal treaty has been concluded between the United States and the United Kingdom. Nor has it been reported which models or testing protocols precisely fall under the change.
The British government also announced new measures on Monday concerning the risks of powerful AI. It wants to use AI in cyber defence, is working on rules for reporting serious AI incidents and is preparing for incidents involving autonomous systems. The government says previous AI pilot projects helped identify and remedy more than a thousand vulnerabilities; that figure comes from the government’s own announcement.
Early access is important to safety researchers because models may change after public introduction or quickly be distributed widely. Testing before launch gives researchers time to report problems to developers. At the same time, such an evaluation remains limited: researchers do not necessarily see all the applications, versions or safety settings that users will later receive.
The development is relevant to European companies and governments because AI models are offered internationally and safety research is increasingly organised across borders. The precise US policy change remains based for now on reporting by Reuters, however; according to the news agency, spokespersons for the White House, OpenAI and Anthropic did not immediately respond.
One story, several perspectives
What is established
- Reuters reported a US policy change concerning British researchers’ access to new AI models.
- The AI Security Institute conducts pre-publication tests.
- The British government announced new AI safety measures.
- The formal status of the US change has not been fully disclosed.
Left
Arguments Powerful AI should be independently tested before entering the market. International safety cooperation and mandatory reporting are needed because, according to this approach, commercial incentives provide insufficient protection.
Values Public oversight, prevention and protection against the concentration of power among technology companies.
Consequences More oversight can expose risks earlier, but may delay launches and raise questions about international jurisdiction.
Centre
Arguments Governments and companies should work together through technical evaluations, clear responsibilities and proportionate rules. Early access is useful, but should not block confidentiality and innovation.
Values Institutional reliability, practicality and a balance between safety and innovation.
Consequences A shared testing framework can increase trust, but requires agreements on information, liability and oversight.
Right
Arguments AI development should not be slowed by broad bureaucratic regulations. Targeted cooperation with expert safety teams is acceptable, provided companies retain room to innovate and national security is protected.
Values Innovation, competitiveness and national sovereignty.
Consequences Fewer rules in advance may deliver faster technological progress, but increase the risk that problems become visible only after introduction.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The text distinguishes the Reuters report from officially confirmed British measures. Unconfirmed details about the US decision-making are presented as such.
- confirmed Reuters reported on 11 October that the US had withdrawn a request to restrict British safety testing. — Reuters describes the change and states that its reporting is based on The Telegraph. source
- confirmed The AI Security Institute tests models before public introduction. — AISI describes pre-deployment access and cooperation with developers. source
- confirmed The British government announced measures concerning AI risks and incident reporting on 12 October. — This is stated in the official GOV.UK announcement. source
- confirmed AISI is part of the British Department for Science, Innovation and Technology. — The official AISI page states its departmental affiliation. source
- confirmed Spokespersons for the White House, OpenAI and Anthropic did not immediately respond. — This statement appears in the Reuters report. source
Editor's note
The US policy change was reported by Reuters on the basis of The Telegraph and had not been confirmed by all parties involved at the time of publication. The British measures and AISI’s role have been officially described by the British government.Sources
- US reverses course on Britain’s testing of AI models, Telegraph says — Reuters via Investing.com
- Britain steps up national defences against growing AI risks — GOV.UK
- About the AI Security Institute — AI Security Institute
More on this in Dutch media
- de Volkskrant — „kunstmatige intelligentie”
- NOS — „kunstmatige intelligentie”
- Het Parool — „kunstmatige intelligentie”