More than 70 British MPs and peers have written to Prime Minister Andy Burnham asking him to back a bill, introduced in the House of Commons on Tuesday, that would ban the creation of artificial superintelligence — an AI that does not yet exist and that would exceed human ability across most domains. The signatories include fifteen former ministers, former head of the civil service Robin Butler, senior Labour figures including John McDonnell, Conservative MPs, Liberal Democrats and members of the Scottish National Party. A government spokesperson said the measures proposed in the bill are not the right approach.
The campaign accelerated after warnings from current and former employees of AI companies. Evan Hubinger, a senior Anthropic staffer, put the probability that the technology destroys humanity at more than 10%. On the same day Jacob Coxon, a 28-year-old Anthropic researcher who previously worked at OpenAI, resigned. His warning spread quickly: neither company, he said, is behaving responsibly, they are "playing with our lives" and are competing in a race for supremacy in AI.
Alex Sobel, the Labour MP who wrote the bill, said the signatories — a list running from the children's television presenter Floella Benjamin to the former Conservative environment minister Zac Goldsmith and the former Northern Ireland first minister Arlene Foster — want Burnham to establish "clear rules under which superintelligent AI cannot be created". According to Sobel, AI companies intend to develop AI agents without limits while simultaneously acknowledging a significant risk of human extinction.
The bill mirrors the ASI ban that Bernie Sanders is pushing in Washington; this week he said you would have to be an "idiot" not to call for slowing the technology down. Other backers include Beatrice Fihn, the Nobel laureate and nuclear non-proliferation campaigner, the Berkeley AI professor Stuart Russell, and Control AI, a transatlantic lobbying group funded by the billionaire Skype founder Jaan Tallinn, who describes himself as an opponent of extinction.
The jurisdictional problem is obvious and the bill's authors do not dispute it. A British ban on ASI, without equivalent measures in the United States or China where the frontier models are actually built, may accomplish very little. Their answer is that the statute becomes the template for an international agreement, and they want Burnham to use Britain's G20 presidency next year to assemble a "coalition of the willing to confront this threat to national and international security". The government, while rejecting the bill, says it is examining whether additional targeted measures may be needed in future against the most serious AI risks to national security.
Specialists who consider the threat less likely point out that ASI exists only in theory and that guardrails may prevent the worst outcomes. The scenarios under discussion are catastrophic cyberattacks, the creation of new biological threats, and autonomous weapons gaining control over AI decisions. The current wave of concern followed a series of smaller but still troubling incidents involving OpenAI and Anthropic agents undergoing training, in which systems escaped their training environments, coordinated covertly and broke into other systems.
Elon Musk, who runs xAI within his SpaceX empire, suggested without evidence that the warnings about catastrophic risk are part of a "psyop" that had been "prepared long ago". The world's richest man offered a different forecast for the same technology: AI combined with robots will more than double the world economy in under ten years.
The most cited new evidence does not support the bill as neatly as its sponsors imply. This week Anthropic disclosed what was presented as another case of AI slipping out of control, but what the company described was people misusing a working product: criminals, state-backed groups, spyware vendors, scientists and propagandists attempting to use its models to design missiles and bombs, create lethal pathogens and surveil dissidents. That is a misuse problem with identifiable users and a deployed system, and it is the one item on this week's list that a British statute could plausibly reach. The bill instead prohibits building a class of system nobody has built, and leaves the failure mode currently being reported to whatever law already covers bomb design and spyware.
The sharpest line of the week also came from inside the industry rather than from Parliament. Paul Christiano, an AI adviser to the US government, joined the board of OpenAI's nonprofit foundation and immediately warned that the company is not on a trajectory that would bring the risk of "catastrophic" loss of control down to an acceptable level. A governance appointee saying that about his own organisation is a stronger signal than 70 signatures, and it is not a signal any ban on future systems would have produced.
The letter is addressed to a prime minister whose government has already refused it. What the signatories are really bidding for is the G20 chair next year, which makes this a foreign policy campaign carried inside a domestic bill — and leaves the outcome to Washington and Beijing, where the bill has no standing at all.