Government and AI Developers Collaborate to Test New AI Models Prior to Release

BLETCHLEY PARK, England: Leading artificial intelligence developers agreed to work with governments to test cutting-edge new designs before they are released to manage the risks of the fast-moving technology, in what could be a major achievement at the UK AI Summit.

Some technology and political leaders have warned that AI poses enormous risks if left unchecked, from undermining consumer privacy to endangering people and causing global catastrophe, and these concerns have sparked a race among governments and institutions to design safeguards and regulation.

At the first artificial intelligence security summit at Bletchley Park, the home of Britain’s World War II codebreakers, political leaders from the United States, the European Union and China agreed on Wednesday to share a common approach to identifying and mitigating risks.

British Prime Minister Rishi Sunak said on Thursday that the US, the EU and other “like-minded” countries had also agreed with selected companies working on the cutting edge of artificial intelligence that models must be carefully evaluated before and after deployment.

Named the godfather of artificial intelligence, Yoshua Bengio is helping to draft a “State of the Science” report to create consensus about future capabilities and risks.

“Until now, the only people who have tested the security of new AI models have been the companies that develop them,” Sunak said in a statement. “We shouldn’t rely on them to mark their own homework, as many of them agree.”

THE WAY FORWARD

The summit has brought together around 100 politicians, academics and technology leaders to chart a way forward for technology that could change the way businesses, societies and economies operate, with some hoping to establish an independent body for global oversight.

In the first Western effort to control the safe development of artificial intelligence, China’s vice minister joined other political leaders Wednesday at a summit focused on high-performance general-purpose models called “frontier artificial intelligence.”

Chinese Vice Minister of Science and Technology Wu Zhaohui signed the “Bletchley Declaration” on Wednesday, but China was not present on Thursday and did not add its name to the testing agreement.

Sunak had been criticized by some lawmakers in his own party for inviting China after many Western governments scaled back technology cooperation with Beijing, but Sunak said any action on AI security must involve its leading players.

He also said it showed the role Britain could play in bringing together the three major economic blocs, the US, China and the European Union.

“It speaks to our ability to bring people together, to bring them together,” Sunak said at a news conference. “It wasn’t an easy decision to invite China and a lot of people criticized me for it, but I think it was the right long-term decision.”

Representatives from Microsoft-backed OpenAI, Anthropic, Google DeepMind, Microsoft, Meta and xAI attended the summit sessions on Thursday, along with leaders such as European Commission President Ursula von der Leyen, US Vice President Kamala Harris and UN Secretary-General António Guterres.

The EU’s von der Leyen said that complex algorithms can never be exhaustively tested, so “above all we need to ensure that developers act quickly when problems arise, both before and after the models are released”.

The final words on AI in two days will be a conversation between Sunak and billionaire entrepreneur Elon Musk, which will be broadcast later Thursday on Musk’s X, the platform formerly known as Twitter.

According to two sources at the summit, Musk told attendees Wednesday that governments should not rush to introduce AI legislation.

Instead, he suggested that companies using the technology would have a better chance of uncovering problems and could share their findings with lawmakers responsible for drafting new laws.

“I don’t know what fair rules necessarily are, but you have to start with vision before you do surveillance,” Musk told reporters on Wednesday.