AI safety – what does this week's China-US summit hope to achieve? And what are the chances?

“May you live in interesting times,” runs a (fake) Chinese proverb, but few would argue that we live in such times. Deep fakery abounds, the Iran war burns like an oil-drum brazier, and the world’s media are debating an AI apocalypse allegedly brought on by some of the least trusted companies in history. A perfect moment, then, for US President Donald Trump – who recently accused AI’s alarm raisers of treason and more – to meet his opposite number, China’s President Xi Jinping. The Chinese premier arrives at Joint Base Andrews just outside Washington, DC, on Wednesday, with White House meetings scheduled for the next day. Xi leaves the US on Friday, with – President Trump has promised – a thick wedge of deals in his neatly-pressed pocket. On the face of it, that outcome seems unlikely. Their previous meeting in Beijing back in May produced little more than an amiable détente between the tech superpowers. Indeed, Kurt Campbell, former US Deputy Secretary of State, told France24 that this week’s meeting is “less a summit and more a spectacle” that is designed for “deep public consumption”. But at least it has given the world’s thinktanks something to talk about. And one such organization, Washington-based CEPA, the Center for European Policy Analysis, has allowed diginomica privileged access to its own Summit pre-briefing. CEPA is generally seen as a trusted, marginally right-of-center non-profit that favors the democratic transatlantic alliances that President Trump has seemed keen to undermine since re-election. So, what has the organization been saying about the Summit, and especially its relevance to the topic of the day: AI safety, and the economic prize for which both nations are racing? Lack of appetite First, neither side has any appetite for collaboration on cybersecurity, says CEPA. Whatever gains the mid-2010s US-China cyber deal achieved have been lost – a point emphasised by the recent Hugging Face and OpenAI breach disclosures. So, on AI, there is little prospect of agreement between the world’s top players. While both seek to triumph economically from the technology – including from Physical AI (robots) – they define risk differently. Beijing's overriding concern is protection of the ruling party, notes CEPA, while the US is more focused on protecting investors. Though as we have seen, it is also keen to protect itself. Kevin Allison is Senior Fellow within CEPA’s Technology Policy Program, and founder and President of Minerva Technology Futures, a policy advisory firm specialising in AI. He says: > One of the big challenges of governments trying to develop frameworks for thinking about these risks is that even among the leading technical and scientific experts, and the leading companies in the industry, there are basic disagreements about how to think about those risks. Or how to think about the balance between the risk and the opportunities. > > > Cybersecurity is a great example. We've had a new wave of headlines about risks, but several years ago, the community of experts was already anticipating that we might reach this point with capable models. So, when there's no agreement on how to think about the risks even among the expert community, it makes it hard to have a good policy discussion. Indeed, President Trump’s online outbursts last week - see diginomica passim - made clear that his administration disagrees with its own AI innovators on the apparent need for a slowdown and better governance. Dr James Lewis (Jim Lewis) is a Distinguished Fellow with the Tech Policy Program at CEPA. He drafted the first comprehensive national strategy for cybersecurity, developing policies for satellites, semiconductors, telecoms, and the internet. He offers a fresh perspective: > This is the summit of the ‘three Ts’: Taiwan, technology, and trade. No one is expecting big breakthroughs – and hopefully there won't be any big explosions. So, it's more how do you manage rivalry? So, we'll see each side press the other for concessions while trying to make as few concessions as possible themselves. I think the US will want to follow up on the commitments China made in the past, none of which have particularly come to fruition. He adds: > But Xi's goals are pretty straightforward. He wants to reassert China's status as a peer to the US. He wants to find ways to see if he can extract concessions for Trump at minimal cost. And he wants to use these to gain an advantage over the US. That is their long-term goal. So, no one has great expectations for this, but hopefully, it won’t fall off the rails. China's dominance of rare earths is the biggest piece of leverage that either side brings to the Summit, believes CEPA. And the November expiry of its export suspension puts a clock on the issue. Political scientist and former US intelligence analyst Elly Rostoum is a Senior Fellow with CEPA’s Tech Policy Program. She explains: > I think probably the very first item on the agenda is what to do with critical minerals. In 2024, we saw that China was willing to flex its muscles and punish the US in response to the tariffs. And the suspension for critical rare earths is coming up in November. That coincides with election season in the US, which means China has huge leverage over the US. They dominate that supply chain, and they have reminded the US that they can do that, so they pretty much have the upper hand on this. Harmful Another issue is AI harms, she confirms: > It's an urgent topic right now in DC. The Frontier AI labs, the Anthropics and the OpenAIs, have come out and said we need help from the government. And some of these labs are even calling for regulation now. Of course, there's no unified voice: NVIDIA said we don't need more regulation, because we already have the laws. > > > Our President has also said we don't need these safe rails, while China has said we shouldn't be afraid of AI threats. We can manage them; we have the infrastructure. But what they have called for is international collaboration. But when they say international, it's really the US, because the only other AI frontier labs are American. Of course, Europe is a huge actor here because they've passed the regulations that touch on this level of governance.” Yet the irony of all this should not be lost on AI developers, or on the wider tech community. In 2023, the UK’s last Conservative Government kicked off the Bletchley Park AI Safety Summit, which was an attempt to forge international partnerships on the issue, with a focus on frontier models. It was followed by a similar event in Seoul. But at the Paris AI Summit last year – the third event in that series – Vice President JD Vance said that the US was no longer interested in AI safety, only in opportunity. Yet here we are today, hearing calls for greater collaboration – from China. Rostoum adds: > A less talked-about area of competition, but a potential area of collaboration, is the biotech sphere. It doesn't get as much publicity, but it's something that CEPA has highlighted. You will be surprised to know that pharmaceutical investments between the US and China rose by $130 billion in just five years. > > > So, that's a huge area of growth, and The Hill [policymakers] is prioritising biotech as an area of competition. The Biosecure Act was passed in December, which pretty much restricted entities that are involved in that with the US government from making deals. They must be approved before collaborating with Chinese companies. Anthropic has come out and said they are worried about bioweapons and the use of AI in the biotech realm. So, I do see that as maybe not a Top Three agenda item, but definitely Top Five. Other technology areas may play a role in the Summit, note CEPA’s analysts. For example, the US position on Chinese electric vehicle (EV) imports probably won't change, since doing so could decimate the domestic auto industry and swing elections in states like Michigan, even as Europe and Canada are more open to cheaper C