‘A critical moment’: concern UK is no up tae speed in actin on AI risks

Towarts the end o Keir Starmer’s time in office, his senior ministers, alarmed bi the latest developments in artificial intelligence, began drawin up plans for a new AI safety law.

They ordered a review o existin legislation tae see whit pouers they already hid, accordin tae thae briefed on the plans, and were explorin whether they cud force the warld’s maist advanced technology companies tae submit their products for safety testin afore launchin them.

In the midst o the poleetical chaos that engulfed the former prime minister as he focht tae remain in office, houever, the plans fell bi the wayside.

Noo, efter twa weeks in which the warld has been tauld there is a greater than 10% chance the technology cud “kill all humans”, and King Charles has warned aboot its potentially destructive capabilities, the government’s attention has turned again tae whit tae dae aboot AI.

Andy Burnham, Starmer’s successor in Doonin Street, has said relatively wee aboot the technology in the past, and in ane o his first acts in office he scrapped the department that hid been leadin the legislative wark.

The prime minister’s focus is on mair immediate domestic problems. But as he grapples wi balancin his first budget and bringin doun the cost o livin, some fear the threat posed bi AI has drapped aff the government’s radar.

“The first duty o government is tae keep fowk safe,” said Chi Onwurah, the Labour MP wha chairs the science and technology committee. “The government’s response tae an existential threat canna be tae thraw its hands up in the air and say there is naething we can dae.”

Breetain has been at the forefront o the debate on AI safety syne Starmer’s predecessor, Rishi Sunak, decided tae mak it ane o his tap priorities in office.

Amid worries aboot whit super-intelligent tools cud dae withoot human involvement, Sunak convened a summit in the UK and persuaded the US, EU, Australie and China tae cooperate on AI safety.

Starmer was mair bullish aboot AI’s potential “tae increase productivity hugely, tae dae things differently, tae provide a better economy that warks in a different way in the future”.

The Guardian has learned that in the former technology department, Emran Mian, the lead ceevil servant, hid plans for AI tae replace almaist every ceevil servant in the department when they retired or moved on tae a new job.

But at the same time, some in government were keen tae remain focused on the risks posed bi the technology.

Shortly afore Starmer’s departure, Yvette Cooper, then the foreign secretary, tauld the Guardian she believed AI posed a Hiroshima-style risk tae humanity.

Starmer’s government took steps tae mitigate the risks Cooper and ithers hid identified. His government allocated £115m in June tae fund a response centre tae deal wi incidents when AI gaes rogue and for a new AI biosecurity programme.

Offeecials in the science department also began reviewin whit pouers ministers hid tae compel companies tae cooperate wi the UK safety regime, and whit else they cud legislate for.

When Burnham took office earlier this simmer, ane o his first moves was tae abolish the Department for Science, Innovation and Technology (DSIT).

Kanishka Narayan, the junior minister in chairge o AI, was gien a promotion and placed in the Cabinet Office, where he continues tae oversee the technology but withoot the logistical support o a dedicated department.

The move alarmed some in the AI industry.

Matt Clifford, the AI investor wha hid advised Starmer and Sunak, posted on X: “Right noo is a critical moment for tech as an economic and national security issue. Tyin up our maist senior science and tech offeecials in a reorganisation wastes time and energy that’s desperately needed for the actual substance.”

Dame Wendy Hall, a computer scientist wha chaired a UK government AI review in 2017, said: “The structure was a bit bloatit in DSIT, but the fragmentation we hae noo has hid a muckle waur impact.”

AI safety rose up the poleetical agenda again ower the simmer.

In July, OpenAI revealed ane o its AI agents hid launched an autonomous cyber-attack on Huggin Face, a real company, triggerin alarm amang technology experts aroond the warld.

Earlier this month, Jacob Coxon, a 28-year-auld researcher, resigned frae his job at Anthropic, claimin the company was “gamblin wi our lives”. His comments were repostit online bi ane o his colleagues, Evan Hubinger, wha addit: “We really dae earnestly believe AI cud kill all humans! I personally think it is >10% within the next decade.”

Just as thae warnings were bein made, the Financial Times revealed that Anthropic hid launched the latest version o its Claude Mythos tool withoot submittin it tae the UK safety institute first.

King Charles intervened this week, convenin a meetin o AI executives in Scotland, where he tauld them: “The task afore you is no merely tae advance technology, but tae ensure that it remains firmly in the service o humanity, community and the naitural warld.”

Public concern is also growin. YouGov said on Friday that just ower a third o Breetons see AI as the third potential cause o human extinction ahint nuclear war and climate breakdown – a figure that mair than doubled syne last year. Twa-thirds think AI has the potential tae end human civilisation, a view shared bi the Nobel-winnin computer scientist Geoffrey Hinton, and ane in sax support a slawdoun in AI development.

In parliament, the Labour MP Liam Byrne has asked the heid o the safety institute, Henry de Zoete, tae testify in front o the business committee he chairs.

De Zoete replied earlier this week, explainin that nane ootside the US hid been allowed tae see Anthropic’s latest product afore it launched and insistin that his institute cud still monitor sic tools efter they hid been released.

“We hae teams o talentit AI researchers workin on research projects based on baith pre-release and post-release testin that are fundamental tae the UK’s understandin o AI risk, across chemical-biological, human influence and loss o control,” he said.

Ministers hae also begun tae comment. Louise Haigh, the de facto deputy prime minister, tauld the TUC congress this week there were “clearly huge risks tae our national security and society if the richt guardrails are no put in place”.

In an article in the Times, Narayan wrote: “Naething is aff the table. Where further regulation, legislation, stranger protections or new pouers are needed, we will act.”

Byrne believes ministers hid a wake-up call.

“I think the government is takin it seriously noo,” he said. “It has become clear in the last twa weeks that perhaps our safety advisers dinna hae the pouer they need tae keep us safe.”

This still leaves the question unresolved bi successive governments – whit can the UK, as a middle-rankin pouer, dae tae curb the micht o some o the warld’s maist advanced and fastest-growin technology companies?

For some, the answer is straichtforrit: regulate and legislate, just as the government wad dae for ony emergin threat.

“The social media companies used tae argue that the government is ower pouerless tae act,” said Onwurah. “But there are things the government can dae. We cud say, for example, that if your frontier model has no gane through the safety institute, we are no gaun tae allow you tae sell your commercial models in our country.

“Or, if your AI gaes and attacks someone else like happened wi Huggin Face, you are legally and financially responsible.”

Mony believe that action will only wark if it is coordinated wi ither countries.

“If the US and China dinna figure this oot, the UK is unlikely tae deliver a safe environment on its ain,” said Byrne. “But a safety regime which is set and enforced bi the UK, Europe, India, Canada and Japan wad be robust.”

The former Cabinet Office minister Darren Jones has called on Burnham tae put the issue at the tap o the agenda when Breetain hosts the G20 summit next year. “The AI safety conversation at the moment is aroond international collaboration on safety testin,” he said. “This is easier tae dae multilaterally, through treaties or agreements.”

Ane former government offeecial addit: “If thegither we say we require you tae pass your models through our safety institutes – if we did that at a lairge enough scale, we wad be able tae stand up tae the US government.”

Whatever the government decides tae dae, it may find that noo is the best time tae dae it.

Unusually, mony in the industry are themselves callin for stricter regulation.

OpenAI’s heid o policy in Europe, Tom Duff Gordon, said this week: “We support stranger UK rules for the handfu o companies, includin OpenAI, developin the maist pouerful AI systems.”

Ane former government offeecial said: “We hae strenths in this country when it comes tae AI. Our AI industry is third in the warld ahint the US and China, and we also hae considerable convenin pouer wi ither countries.

“The question for me is whether the new prime minister is alive enough tae this problem tae tak advantage o thae strenths.”