Blow for Sunak’s AI summit as leaked document shows UK scaling back research plans

LONDON — Prime Minister Rishi Sunak’s hopes of securing international agreement for a new scientific body focused on risks from the most powerful AI systems have been dealt a blow, according to a final summit communiqué obtained by POLITICO.

While nations participating in the United Kingdom’s AI safety summit will agree that frontier AI could cause “serious, even catastrophic, harm,” they are expected to signal that further scientific study of the issue could be carried out through existing efforts, including the United Nations and Global Partnership on AI.

That change will be seen as a setback to the U.K. government, which had hoped to establish a new global AI research body at its flagship AI Safety Summit on November 1 and 2.

The document, dated October 25, was presented by the General Secretariat of the European Council to EU member states. It is described as the “final version of the communiqué.” A schedule seen by POLITICO stated that Wednesday October 25 was the final deadline for agreeing the communiqué.

The reference to a network that “encompasses and complements” existing efforts is a change from an earlier draft of the statement, also obtained by POLITICO. The previous draft only said the network should “support” multilateral collaboration, and did not reference the U.N.

The U.N. announced the formation of a high-level advisory body on AI in August and was expected to name appointees on Thursday. It is due to present a final report by September 2024.

A U.K. official with close understanding of the summit said they were still negotiating edits to the document and disputed that the new wording reflected a scaling back of their ambitions.

“I genuinely think it’s the opposite,” said the official who was granted anonymity to discuss multilateral negotiations. “At least that’s the intention based on feedback we’ve had.”

In another potential setback for the U.K., an earlier reference to “future international AI Safety Summits” is now in square brackets, suggesting that governments have not reached agreement. The U.K. technology secretary said on Tuesday she hoped the summit at Bletchley Park would be the “first of many.”

In other changes, a reference to the UNESCO’s Recommendation on the Ethics of AI and the G20 have been deleted. Though the reasons are unclear, China is a member of both bodies.

The final communiqué also adds that proportionate governance policies “could include making, where appropriate, classifications and categorisations of risk based on national circumstances and applicable legal frameworks” and notes “the relevance of cooperation, where appropriate, on approaches such as common principles and codes of conduct.”

Another document, also obtained by POLITICO and attributed to the General Secretariat of the European Council, states that feedback from an EU working group to the earlier draft statement sought to align it “even more closely with the approach to AI being pursued by the Union, notably as reflected in the work on the AI Act.”

This story has been updated to include the comment from the U.K. official.