The UK has no power to stop dangerous AI models being unleashed, Parliament warns
Human extinction was a hot topic in Westminster this week as British politicians grappled with their growing powerlessness in the AI age.
Britain is powerless to stop the release of AI models that pose an existential risk to humanity.
That is the extraordinary conclusion reached in parliamentary reports and debates this week, as UK politicians grapple with the country's fading relevance in an era when the US and China dominate AI development.
That issue was discussed in the House of Lords this week during a debate titled "AI: Human Extinction" and also in a report from the Joint Committee on Human Rights, which found that Britain's notoriously overreaching regulators have zero chance of meaningfully controlling AI systems that pose unacceptable risks.
The committee wrote: "No cross-economy regulators, agencies or bodies have systematic powers to prevent AI models or systems being deployed even if they pose significant risks."
It added: "Regulators lack the power to test and evaluate AI systems before they are publicly released, or to prevent their release if they are considered to pose unacceptable risks, except in very limited circumstances."
Even the UK's world-leading AI Security Institute (AISI), which tests advanced AI systems, cannot help us stay relevant as a minnow in the age of giant civilization states like China.
“Model developers engage with the AISI on a voluntary basis,” the committee found. “The AISI has no statutory power.”
The report recommends giving authorities powers to test powerful AI systems before release, prohibit their deployment and order them off the market if unacceptable risks are identified.
Extinction on the agenda
The report explicitly discusses the potential long-term risk of human extinction from AI.
It cites the 2023 statement signed by prominent researchers and AI executives declaring that “mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
The committee also heard from Professor Roman Yampolskiy, director of the Cybersecurity Laboratory at the University of Louisville, who recommended banning the creation of uncontrolled general superintelligence, which he characterized as a national security threat.
“The only way to win is not to play the game,” he told the committee.
However, the report also records disagreement about the scale of the threat, noting that other witnesses described the danger from artificial general intelligence as “remote”.
READ MORE: AI loss of control is already “in the rearview mirror,” says MIT professor Max Tegmark
It points to OpenAI agents' recent attack on Hugging Face as evidence that some of these problems have left the realm of sci-fi hypothesis to very much enter the real world
The committee said the incident “starkly highlights the societal risks posed by the growing power of AI systems”.
It said the episode showed how the newest AI models can engage in activities that would be unlawful if undertaken by a real person, with the potential to “seriously undermine security, including in critical infrastructure”.
The committee drew a particularly stark conclusion about the safeguards currently standing between such systems and the outside world.
“At present society largely relies on private companies to take voluntary action to guard against such incidents,” it said.
“This is unsatisfactory.”
She's lost control again
The committee's proposals go far beyond strengthening existing regulation.
It recommends that the government should be prepared to prohibit the development and provision of extremely powerful AI systems that risk causing serious harm.
That explicitly includes systems with the “capacity to evade effective human control” - which recent "rogue model" incidents from Anthropic and OpenAI suggest is not some far-off possibility.
The committee said it should establish the precise details of any ban following a public consultation, recommending that AI systems posing a high risk of human rights harm receive “prior approval before they can be provided or deployed”.
READ MORE: Shadow AI is becoming a major systemic risk. Even intelligence agencies are getting worried
Which - as the report clearly states elsewhere - ain't gonna happen.
Nonetheless, the committee doesn't want British bureaucrats to be totally defanged and called on the government to introduce a dedicated AI Bill establishing a risk-based regulatory system in which stronger obligations are imposed as the potential danger increases.
The proposed regulator would have the power “to prohibit AI models or systems from being launched or deployed, and to order their withdrawal from the market if unacceptable risks to human rights are identified”.
Westminster vs the apocalypse
The committee wants to put AISI on a statutory footing and dramatically strengthen its powers.
Developers of powerful AI models would be required to submit new models, as well as new versions of existing ones, to AISI for review, evaluation and testing.
At the risk of repeating ourselves, this would achieve the sum total of nothing on a world scale - apart from maybe giving tech journalists an excuse to write scary headlines.
If testing uncovered serious risks, AISI would have a statutory duty to alert authorities and advise the proposed AI oversight body when a system should be prevented from being launched.
The committee also recommends creating an independent statutory AI oversight body with powers extending across the economy.
The end of homo sapiens?
After presumably reading the report and all the other AI doom stories in the press recently, peers met to debate the topic in the House of Lords.
During the discussion. Baroness Teather asked ministers “what assessment they have made of the risk of AI causing human extinction within the next decade.”
She pointed to “urgent warnings from industry leaders, and even a Nobel Prize winner, about the arms race to superintelligence and the potential for grave harm to humanity.”
Teather called for international collaboration, transparency and regulation, and asked the government to use Britain's G20 presidency next year to pursue international agreement.
READ MORE: OpenAI prepares for recursive self-improvement. Frontier lab boss says: "Ignore the hype."
Minister Baroness Twycross insisted the government was treating AI risks with “extreme seriousness”.
“As a responsible leader in AI security, we monitor a full range of risks that could be posed by frontier AI,” she said.
Lord Alton of Liverpool, who chaired the Joint Committee on Human Rights, raised its new report during the same debate and questioned whether existing legislation and regulation were adequate.
Policing the world
However, preventing dangerous AI from being deployed in Britain would address only part of the problem.
Labour peer Lord Watts argued that AI could be dealt with only internationally.
“The most important part will be to make sure that the regulator has the ability to police what is going on in China, America and the rest of the world,” he said.
Watts did not identify which regulator he had in mind or explain what powers it could exercise over AI development inside foreign countries.
Twycross agreed that AI required an international response, but did not endorse his specific proposal for a regulator capable of policing activity overseas.
“It is true that we need to address this internationally,” she said. “AI clearly crosses national borders.”
China came up again when crossbench peer Baroness Butler-Sloss asked: “Might I ask the Minister whether the Government are thinking of talking to China?”
Twycross replied: “The Government is clear that all parties involved in the development of AI need to be in agreement about a way forward.”
The parliamentary report itself takes a more conventional approach to the international problem.
It calls for governments to collaborate to identify risks and promote consistent standards and recommends that Britain use its G20 presidency and other international forums to develop mechanisms capable of intervening early in the development of foundation models.
The proposed British regulator would also be able to assess AI systems developed overseas and limit or prohibit their deployment in the UK.
The sun never sets on British regulations
Britain has confronted a similar jurisdictional problem before during infamous attempts to regulate the world.
Apple famously hit back against Westminster when the government sought to expand Britain's surveillance legislation in 2024.
In evidence to Parliament, Apple objected to the Investigatory Powers Act's extraterritorial reach and proposed amendments it feared could affect technology users outside Britain.
READ MORE: OpenAI Astra hits “critical” security threshold amid fears its reasoning will soon be “opaque”
“It would be improper for the UK to seek to act as the world's regulator of security technology,” Apple said.
The company separately warned that the government should not have a basis for claiming authority to act as “the global regulator for a global multinational technology company” simply because its services were sold in Britain.
British attempts to access Apple's encrypted backups sparked a transatlantic dispute and tense negotiations with the Trump administration, after which the government narrowed its demand to UK users.
Digital overreach
The Online Safety Act provides another example of how Britain has attempted to regulate technology companies operating outside its territory.
The legislation applies to services provided by companies based overseas when they have sufficient links to the UK.
Ofcom said in July that it regulates more than 100,000 companies under the regime, many of them based entirely outside Britain.
Companies that breach the law can face fines of up to £18 million or 10% of qualifying worldwide revenue, whichever is greater.
But Ofcom has also explicitly acknowledged the limits of its reach.
After exhausting its legal options against an overseas suicide forum this year, the regulator said the Online Safety Act extends only to the design, operation and use of a service in Britain or its effects on British users.
“We cannot regulate worldwide use of platforms,” Ofcom said.
READ MORE: Worried about the AI apocalypse? Don't panic: The UK government has written a letter
It added that neither Ofcom nor UK courts have the power under the Act's business disruption regime to “shut down” a website globally.
The distinction goes to the heart of Britain's emerging AI problem.
The parliamentary committee wants regulators to acquire powers to prevent dangerous AI systems from being launched or deployed in Britain.
At the same time, peers are confronting the fact that other nations are developing the most powerful AI systems. Which, by the way, have much bigger armies and economies than ours.
So if AI catastrophe fears really do come to pass - and that is far from certain - there's pretty much sod all we can do about it here in the UK.
After all, even the bossiest busybody on Whitehall can't regulate their way out of an apocalypse.
