UK’s AI safety summit gets thumbs up from tech giants

Make way for another forum on AI safety. The UK government has announced it will convene a “global” AI summit this fall with the aim of agreeing “safety measures to evaluate and monitor the most significant risks from AI”, as its PR puts it.

There’s no word on who will attend as yet — but the government says it wants the discussion to feature “key countries, leading tech companies and researchers”.

“The summit, which will be hosted in the UK this autumn, will consider the risks of AI, including frontier systems, and discuss how they can be mitigated through internationally coordinated action. It will also provide a platform for countries to work together on further developing a shared approach to mitigate these risks,” it adds.

Prime minister Rishi Sunak is in the US today where, per the government line, he will meet with president Biden and press for “joint leadership” of technologies such as AI, among chat on other economically significant issues.

Notably the press release announcing the UK’s ambition to host a global AI summit simultaneously bundles a separate claim, vis-a-vis “global companies expanding their AI work in the UK”, with the government spotlighting developments such as OpenAI opening a London office last week.

The PR is also dominated by canned quotes from tech giants and AI firms, with the likes of Google DeepMind, Anthropic, Palantir and Faculty lavishing praise on the summit plan via supporting statements from senior execs.

(For a flavor of the industry flattery embedded in the government’s PR, DeepMind’s Demis Hassabis proclaims Sunak’s “Global Summit on AI Safety will play a critical role in bringing together government, industry, academia and civil society”; Anthropic’s Dario Amodei commends the PM for “bringing the world together to find answers and have smart conversations”; and Faculty’s Marc Warner suggests the UK is “perfectly placed” to provide “technological leadership” and “foster international collaboration” — so, er, pass the sick bucket… )

The strategy the UK appears to be plumping for here is to position itself as the AI industry’s BFF (or, well, stooge) — in a way that could work against existing international efforts to agree meaningful guardrails for AI if it ends up driving a wedge between the US side and other international players.

The summit announcement comes about two weeks after Sunak met with a number of tech execs helming AI giants, including Anthropic’s Amodei, DeepMind’s Hassabis and OpenAI’s Sam Altman. After which the government suddenly started squawking about existential AI risk, in a clear parroting of the sci-fi concerns AI giants have been promulgating vis-a-vis non-existent “superintelligent” AI systems — in a bid to frame the debate about AI safety by zeroing in on theoretical future risks — while downplaying discussion of actual harms being caused by AI in the here and now (such as privacy abuse, bias, discrimination and disinformation, copyright infringement and environmental damage, to name a few).

In another sign of the lavish AI industry love-in for UK Plc rn, Palantir’s CEO Alex Karp was interviewed on AI by BBC Radio 4’s Today program this morning where he made a point of heaping praise on the UK’s “pragmatic approach to data protection”, as he put it — going on to compare the UK’s under-enforcement of privacy rules favorably to the EU’s more robust enforcement of the General Data Protection Regulation (which, by contrast, quickly forced ChatGPT to provide users with more information and controls), as well as claiming it will “be much much harder for the continent to come to terms with large language models [than the UK]”.

It remains to be seen what the Biden administration will make of Sunak’s AI safety summit. Or, indeed, whether anyone of significance from the US government will attend. But AI giants being mostly US-based certainly muddies the AI regulation conversation over the pond.

US lawmakers remain concerned about the burden of AI regulation on industry — and are demonstrably more reluctant to rush in with guardrails than, for example, their counterparts in the European Union.

As a third country to both those sides, the UK has a choice to make over where to throw its hat on international AI rules. All the signs are it’s aiming to try to use this topic — and US AI giants — as a strategic lever to ratchet itself into a closer relationship with the US, based on aligning over more dilute AI rules (assuming the US agrees to play this game).

The UK is actually a late convert to the discussion on how to regulate AI. Only a few months ago it put out an AI white paper saying it didn’t see the need for any new bespoke rules or oversight bodies for AI — preferring to load the responsibility onto existing over-worked regulators (without expanding their budgets) by asking them to devise and issue context-specific guidance. The name of that whitepaper? “A pro-innovation approach to AI regulation.”

It’s also making this AI summit move at a time when governments, regulators and lawmakers around the world are already responding to rising alarm about the safety risks flowing from fast developing machine learning technologies by mobilizing a variety of discussion tracks and initiatives with the goal of clinching international agreement on safeguards and safety standards.

The OECD already adopted AI principles all the way back in May 2019. While the FTC put out AI guidance in April 2021. And even the US Department of Commerce’s National Telecommunications and Information Administration (NTIA) started consulting on how to boost AI accountability this April. The UN is also looking at AI.

Then there’s the G7 leaders’ “Hiroshima process” — a recent track comprised of cabinet-level discussions between G7 countries on AI governance which is due to report by the end of the year. While, before that, G7 countries and others launched the Global Partnership on AI — which is aimed at promoting responsible, human-centric development and use of AI technologies by sharing research and foster international collaboration toward trustworthy AI.

The European Union, meanwhile, presented its own draft legislative for regulating AI over two years ago. The bloc’s lawmakers are now busy hammering out agreement on a final text of that framework — including considering how it should tackle generative AI — with political agreement on the EU AI Act sought for by the end of this year. (Although the pan-EU law won’t be in force for several years after that.)

The EU and US are also working (or at least talking) together on an AI Code of Conduct which is being conceived as a stop-gap set of voluntary standards until legislation comes in — doing so via a transatlantic talking shop, called the US-EU Trade and Technology Council (TTC), a forum the UK is not party to having left the EU following the Brexit referendum.

Last week the EU said it would begin drafting this AI Code of Conduct, adding it hoped to have something on paper within a matter of weeks.

Although it was less clear after the TTC meeting how much buy-in the US side was committed to. But US lawmakers were in the room talking.

Discussing the AI Code in a briefing with journalists last week, EU EVP Margrethe Vestager, who heads up the bloc’s digital strategy, underscored how this EU-led initiative could, very quickly, be moulding global AI guardrails, telling journalists: “If we can start drafting with the Americans, the rest of G7, invited guests and have industry sign up for it — of course also for us with some third party validation — then we could cover one-third of global population within a very, very short timespan. And that may be a good thing.”

So the bloc is clearly working at pace to seize the opportunity to apply the ‘Brussels effect’ to first-order global AI rules.

EU lawmakers also recently announced a second track on AI regulation: An AI Pact which is aimed at getting the tech industry to agree to voluntarily compliance with provisions coming down the pipe in the aforementioned EU AI Act before that legislation is in legal force (which likely won’t be before 2026, per Vestager).

So the EU is all over the AI regulation piece from multiple angles (hard law, industry code, international code and a key participant in other global initiatives).

The US is also certainly involved in the conversation — albeit while holding itself at a cooler distance and keeping its cards closer to its chest, likely on account of US AI giants’ lobbying.

By comparison, the UK is just pulling on its trousers as it scrambles to catch up. (After its PR kicks off with an eye-catching claim that the UK will host “the first major global summit on AI safety” it goes on to pay lip-service to some of the existing international efforts on AI safety, over the past several years — making a passing mention of “recent discussions at the G7, OECD and Global Partnership on AI” which it implies the UK summit will build on. There’s no mention of the EU’s AI multi-pronged efforts; instead there’s a Brexit flavored ‘porky’ claiming: “Our departure from the EU also allows us to act more quickly and agilely in response to this rapidly changing market.”)

How significant the UK siding with US AI industry on self-serving safety rules might be remains to be seen. But there is an opportunity for Sunak to drive a wedge in EU efforts to bring the Americans into a more robust international AI Code. Hence all the AI industry flattery.

Following the TTC meeting between EU and the US lawmakers, an EU source told TechCrunch the discussions revealed divergent approaches — saying that the US does not seem ready to regulate, whereas they said the bloc remains committed to its established view that a responsible rollout of AI requires robust and ambitious regulation.

We were also told there is concern on the EU side that discussions on global AI principles do not end up getting watered down to a lowest common denominator that could end up undercutting the risk-based AI Act legislative framework the bloc has in train.

So there’s perhaps more at stake and riding on Sunak’s AI Summit than might seem at first blush.