Leadership1 publisher2 min readPublished
Burnham skips compulsory AI testing pledge, unveils US-UK defence AI partnership
The leaders of Australia, Canada, Germany, the EU, Bahrain and Singapore have signed Finland's call for compulsory pre-release testing of frontier models. The UK has not, and its offer at the UN is a defence AI partnership with Washington.
The Board Room · Leadership desk
What happened
- Downing Street briefed that AI would be a main theme of Burnham's New York visit, with his Tuesday evening UN general assembly speech built around harnessing the technology to drive down the cost of living.
- Sky News's Sam Coates reported that many leading countries and the EU backed a Finnish government call for compulsory safety testing of frontier AI models before release, and that the UK did not sign up.
- Darren Jones, chief secretary to the prime minister under Keir Starmer, urged the government to sign the Finnish proposal and to offer to broker agreement with the United States on how it would apply.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- contradiction Downing Street presents Burnham as driving forward global work on AI. Coates reports the UK declining the one multilateral pre-release testing commitment its closest allies signed, and calls the safety language in the press release watery.
- constraint A compulsory-testing commitment would bind the UK to require something it cannot impose on American-developed models by itself. Jones's answer was to sign first and negotiate the application with Washington.
- capability Deterrence work that needs American models now has a named route to them: the MoD's Rapid AI Delivery Taskforce and the US Department of War's Chief Digital and Artificial Intelligence Office are the counterparties. UK infrastructure monitoring can be built on jointly developed systems.
Darren Jones, chief secretary to the prime minister when Keir Starmer was in Downing Street, set out both sides of the choice in public. He said the signatories to the "Call for Control of Frontier AI Models" "are the United Kingdom's natural allies" [11]. He also said: "The UK has special access to American AI models through the AI Security Institute, which is partly why the UK Government will want to remain in lock step with the United States" [12]. One sentence is about alliances; the other is about an access arrangement the UK holds with American developers. Jones proposed doing both: "I encourage our government to do so and to offer to broker agreement on its application with the United States," he said [13].
What Burnham took to New York is a defence programme. Downing Street says the new AI and Autonomy partnership will help the UK protect critical national infrastructure, from sea lanes to airspace, using technology developed on both sides of the Atlantic [4]. The press notice pairs the Ministry of Defence's Rapid AI Delivery Taskforce with the US Department of War's Chief Digital and Artificial Intelligence Office [5]. The Guardian's live blog said the press release implies Burnham is more interested in how AI can be used to threaten Russia than in steps to stop AI wiping out humanity [17]. Coates called the language in the press release "much more watery" [9].
Anyone timing a frontier release should be clear about what the record contains. Coates reported the call's date and six named backers: Australia, Canada, Germany, the EU, Bahrain and Singapore [7]. The report says nothing about what those signatories would require of a developer, or when [21]. Five of the six are national governments [19], and each would have to write its own version into law before a test became a condition of shipping. The UK stayed off the call, and it already has special access to American models through the AI Security Institute [6][12].
A leaders' call published the day before the general assembly opens changes nothing about what any lab ships this quarter. The part that carries into next year is who drafts the definitions. Coates reported Burnham arguing that Britain is uniquely placed to lead international AI governance, mediate between the US and China and make AI a centrepiece of its 2027 G20 presidency [8]. No 10 has confirmed that Manchester will host next year's G20 summit [15]. The call is dated 21 September; the UK's response, a day later, was a different announcement [20].
What to watch
- Whether the UK signs the Call for Control of Frontier AI Models late, or publishes its own pre-release testing standard before the Manchester G20.
- Scope, funding and command arrangements for the UK-US AI and Autonomy partnership beyond the pairing of the MoD taskforce and the US CDAO.
- Whether any signatory turns the call into a domestic pre-deployment requirement with thresholds and dates attached.