MPs in the UK want emergency powers that would let the government shut down data centres and AI systems during a national emergency. The proposal is in an amendment to the Cyber Security and Resilience Bill, according to Computer Weekly. The amendment says ministers could intervene if an AI system created a “catastrophic risk” to essential services, national security or human life.
The proposal came from Labour and Co operative MP Alex Sobel and received backing from 11 MPs. Control AI, a campaign group calling for stricter AI regulation, also backed the amendment. Computer Weekly reported that the proposal would give the Secretary of State emergency powers to turn off data centres or AI systems judged to present a serious risk.
What Does This Proposal Cover?
The proposed powers cover cyber attacks, hostile state activity and autonomous AI systems capable of attacking computer networks. Computer Weekly also reported that the amendment mentions “super intelligent AI” that could operate without human oversight.
Andrea Miotti, founder and CEO of Control AI, wrote on X, “It also helps in cases where the government has reason to believe super intelligent AI, AI that can autonomously compromise national security, is being developed on UK soil.”
Miotti also wrote, “The UK is not truly sovereign on AI if it can’t pull the plug when AI national security threats happen on its soil.”
Why Are MPs Discussing This Now?
The amendment follows mounting attention around frontier AI models. Computer Weekly reported that frontier systems can identify unknown software vulnerabilities and find ways to exploit them. That creates risks for essential services and cybersecurity systems.
Computer Weekly reported that Anthropic recently released its Claude Mythos model to selected technology companies through Project Glasswing. Anthropic claimed the model discovered thousands of security vulnerabilities, including flaws that had gone unnoticed for years.
The amendment also proposes emergency exercises for data centre operators. Operators would need technical systems capable of complying with government shutdown orders. After an incident, data centres would need monitoring and mitigation measures before resuming operations.
The proposal also gives data centre operators legal options. Computer Weekly reported that operators would receive notice before shutdown action wherever possible, and they could challenge decisions in the High Court.
What Does This Mean For Startups?
For startups building AI products, the proposal introduces an entirely new regulatory reality. Newer companies often move faster than large corporations, though emergency shutdown powers mean founders may need stronger oversight systems much earlier in development.
Nik Kairinos, CEO and Co Founder of RAIDS AI, said, “This proposal rightly recognises that some AI failures could have consequences beyond a single company or sector, particularly where critical infrastructure, cybersecurity or national security are involved.”
Kairinos also said, “While a ‘kill switch’ might sound dramatic, it reflects a growing understanding that once increasingly intelligent AI systems are live, it’s vital that we have a way to regain control if they begin behaving unpredictably or dangerously.”
Startups working in cybersecurity, infrastructure, defence or automation may receive closer examination from regulators and data centre operators. Investors may also expect stronger monitoring systems before funding advanced AI products.
Kairinos said, “However, emergency shutdown powers should be the last resort, not the main safety strategy. If an AI system has reached the point where it needs to be switched off entirely, the warning signs have already been missed.”
Could The Proposal Change How AI Companies Operate?
The amendment has not received government approval, though the proposal already gives AI companies a preview of future regulation. Startups may need stronger auditing systems, emergency response procedures and real time monitoring much earlier than expected.
Kairinos said, “Organisations therefore cannot rely on a government switch to manage their AI risk, and must take ownership of the actions of their AI systems.”
He also said, “That is why continuous monitoring matters, because it gives organisations the ability to detect problems as they emerge, respond in real time and adapt as systems, use cases and threat landscapes change.”
Kairinos ended off with a warning about delayed responses. He said, “Without that visibility, organisations may only discover unsafe or rogue behaviour once the damage has already been done.”
