UK regulators lack power to stop AI risks, politicians say
AI Security Institute has no authority to compel companies to share models
British politicians have called for new legislation to strengthen AI oversight, warning that regulators lack the powers to prevent the release of AI systems that could pose serious risks to human rights and national security.
The Joint Committee on Human Rights (JCHR) said the existing regulatory framework was fragmented and largely focused on AI once it had already been deployed.
It has instead recommended a new risk-based system, under which the most powerful or potentially harmful AI systems would face stricter requirements before being released.
The committee said the AI Security Institute (AISI), the government's main body for testing advanced AI models, is a research organisation rather than a regulator. As such, it lacks the statutory authority to compel companies to submit models for testing or to prevent their release.
The government has said AISI's voluntary arrangements with leading AI companies have allowed it to identify dozens of vulnerabilities that developers addressed before releasing models.
Calls for tougher checks on powerful AI
The committee's report said high-risk AI systems should require prior approval, with obligations imposed throughout the technology's development and deployment.
It also called for some uses of AI to be prohibited, including certain applications involving subliminal techniques and inappropriate profiling or biometric data.
Systems capable of causing catastrophic harm on a wide scale should also be prohibited, with the committee citing artificial general intelligence and superintelligence as examples.
Alex Sobel, the JCHR's Labour chair, said the UK was not prepared for the potential consequences of rapidly advancing AI.
"Nowhere in the world, including the UK, has a current legislative and regulatory approach to AI that is fit for purpose," he said.
The committee wants people to be told when AI is being used to make decisions that significantly affect them. It also wants stronger rights to challenge such decisions and obtain redress when things go wrong.
It proposed creating a single independent AI oversight body with powers to investigate complaints, set codes of practice and transparency requirements, and sanction organisations that break the rules.
Ministers urged to heed AI safety warnings
First Secretary of State Louise Haigh said the government should "heed the warnings" from people working at the forefront of AI development.
Speaking at the TUC conference, she said AI could transform public services, strengthen businesses and contribute to scientific breakthroughs, but warned of "huge risks to our national security and society if the right guardrails are not put in place."
Her comments followed warnings from researchers at AI companies Anthropic and OpenAI about the potential for advanced AI to cause catastrophic harm.
Anthropic and other technology companies have also been involved in renewed discussions about whether the pace of AI development should be slowed.
Labour MP Darren Jones has called for international cooperation on AI regulation, including using next year's UK-hosted G20 summit to discuss common rules.
TUC general secretary Paul Nowak said AI represented a serious threat to workers and wider society and called for international action.
Trade union representatives at the conference also raised concerns about the growing use of AI to monitor workers and assess performance, as well as fears that automation could reduce employment in some sectors.
Haigh said workers should share in the benefits when AI increases productivity, rather than the gains going solely to shareholders.
Meanwhile, debate has intensified over proposals for an AI "kill switch" that could allow an external body to halt particularly dangerous systems.
The Cabinet Office on Friday rejected calls for such a mechanism, saying that blocking access to AI models in the UK would not necessarily stop them from being developed or misused elsewhere.