Britain is not sovereign without an AI kill switch
Guest Article: Alex Sobel MP writes about why the UK needs an AI kill switch to remain sovereign and how his amendment to the Cyber Security and Resilience Bill would provide just that.
Just weeks ago, the AI company Anthropic shocked observers by deciding to withhold wide-release of their newest AI model Mythos because of the dangerous cyber capabilities it displays. The company said Mythos poses a national security risk and reported that the model found vulnerabilities in all major web browsers and operating systems, including some that cybersecurity experts had not found for decades.
Make no mistake: advanced AI is redefining the cybersecurity landscape and, with it, the national security threats AI brings to bear on our critical infrastructure. The UK is woefully unprepared to meet this moment.
We have no legal framework for responding to AI-driven cyberattacks at scale, no protocol for shutting down dangerous AI systems or data centres in the UK, and no clear authority empowered to act in an AI emergency. A coordinated cyberattack on critical infrastructure would be devastating, and superintelligent AI vastly smarter than humans could match a nuclear strike in the damage it does. We are nowhere near ready for either.
That’s why I have proposed an amendment adding ‘kill switch’ powers to the Cyber Security Bill currently being considered by Parliament. This amendment would give the Government the power to direct the shutdown of data centres or AI systems deployed in the UK in the event of AI security or operational emergencies.
A ‘kill switch’ is a common sense preparedness mechanism, and even Washington recognises the danger of AI we cannot shut down. In April, President Trump, when asked whether the government should have a ‘kill switch’ for AI technology, said, “There should be”. This amendment would empower the Government to respond to AI-related emergencies on UK soil, covering two distinct threats: AI-driven cyberattacks and the development of superintelligent AI.
Consider what will happen when a data centre hosting critical services for the NHS, financial institutions, or energy infrastructure is compromised by an AI-driven attack spreading faster than response teams can contain it. As we witness the dawn of highly capable, autonomous AI hacking, such scenarios are no longer hypothetical.
Yet the current regulatory framework leaves the Government without tools to act decisively in the wake of large-scale cyberattacks, whether orchestrated by a rogue state, criminal actors, or autonomous AI systems themselves taking over the UK’s computing resources. As the MI5 Director General warned in late 2025, we need to scope out “potential future risks from non-human, autonomous AI systems which may evade human oversight and control.”
Right now, the UK does not have the sovereign capability to ‘pull the plug’ in the case of dangerous AI cyberattacks or the takeover of Britain’s computing infrastructure. My amendment would change that. This ‘kill switch’ would be a last-resort tool for the Government to pull the plug when things go wrong, or when there is sufficient evidence that things will go wrong.
The amendment also helps address the development of superintelligent AI using UK-based data centres. Superintelligent AI would be vastly more capable than humans and able to irreversibly escape human control, which is why leading AI experts warn that it would pose an extinction risk to humanity.
Currently, even if there was evidence that superintelligent AI systems that could autonomously compromise national security were being developed in Britain, there is no clear protocol for the Government to intervene and proactively stop the threat. This kill switch amendment would thus enable the Government to direct the shutdown of the given AI system, and the data centre housing it, if they assessed that an actor was developing superintelligent AI using UK data centres.
Concerns about the security risks posed by superintelligence are shared widely across Parliament. Over 100 cross-party parliamentarians have joined me in backing a statement from ControlAI warning that superintelligent AI systems would compromise national and global security. It’s time for the Government to start treating the development of superintelligent AI like the unprecedented national security threat that it is, and this amendment is a concrete move in that direction.
Establishing this kill switch power would be a vital first step in preparedness, but this amendment alone will not be enough. Appropriately responding to the threats posed by advanced AI will require preventing the development of superintelligent AI and launching robust international coordination. As I have called for alongside the ‘godfathers of AI’ and 800 other prominent figures, we ultimately need to reach a global prohibition on developing superintelligent AI due to the extinction risk it would pose. Fundamentally, we must ensure that AI does not threaten our human right to life, which is why I’m working with colleagues on the Joint Committee on Human Rights to conduct an inquiry into the current and future human rights risks posed by AI.
With models like Mythos already reshaping the threat landscape, the UK cannot afford to build its response mechanism after the fact. Britain is not truly sovereign as long as we are helpless to act in cases of an AI emergency happening on our own soil. This ‘kill switch’ amendment is that mechanism. Parliament should secure it now, before it is needed.
Alex Sobel is the Labour and Co-operative Party MP for Leeds Central and Headingley (@alexsobel on X/Twitter). He has been a member of the UK Parliament since 2017. His AI kill switch amendment has also been covered in The Telegraph.
 | A guest post by
|
I rise today to speak in strong support of the amendment before the House — an amendment that would grant the Government the power to direct the shutdown of dangerous AI systems or compromised data centres in the event of an AI‑related security emergency.
This is not a theoretical debate. It is a matter of national security, public safety, and sovereign control over the digital infrastructure upon which modern Britain depends.
Just weeks ago, the AI company Anthropic took the unprecedented step of withholding wide release of its newest model, Mythos, after discovering that it possessed advanced cyber‑offensive capabilities. Mythos was able to identify vulnerabilities in every major web browser and operating system — including weaknesses that cybersecurity experts had not uncovered in decades. The company itself warned that the model posed a national security risk.
If private developers are now producing systems capable of discovering zero‑day vulnerabilities at scale, autonomously, and at machine speed, then we in Parliament must confront a simple truth: our current legal framework is not built for the world we have entered.
At present, the United Kingdom has no statutory mechanism to shut down a dangerous AI system.
No emergency protocol.
No designated authority.
No power to compel the shutdown of a compromised data centre hosting critical national infrastructure.
Madam Deputy Speaker, that is a sovereignty gap.
A security gap.
And a preparedness gap.
Our hospitals, our financial institutions, our energy grid, our transport networks — all rely on cloud‑based systems that could be compromised by an AI‑driven cyberattack spreading faster than any human response team could contain. The consequences of such an attack would be devastating. And as the MI5 Director General warned in 2025, we must prepare for “potential future risks from non‑human, autonomous AI systems which may evade human oversight and control.”
Yet today, if such a system emerged on UK soil, the Government would be powerless to act.
This amendment changes that.
It provides a last‑resort kill‑switch power — a sovereign safeguard enabling the Government to intervene decisively when an AI system threatens national security or when a data centre becomes the vector of an uncontrollable cyber incident.
Let me be clear: this is not a tool for routine use.
It is not a tool for political convenience.
It is a tool for emergencies — for the moments when delay would mean disaster.
And it is not only about today’s threats. It is about tomorrow’s.
Leading AI researchers, intelligence officials, and over 100 cross‑party parliamentarians have warned that the development of superintelligent AI — systems vastly more capable than humans and able to escape human control — would pose an unprecedented national security risk. If credible evidence emerged that such a system was being developed in a UK data centre, the Government currently has no clear authority to intervene.
This amendment closes that gap. It ensures that the United Kingdom retains the sovereign ability to act before a threat becomes irreversible.
Madam Deputy Speaker, Parliament has a responsibility to anticipate danger, not merely react to it. We cannot wait for an AI‑driven cyberattack to cripple the NHS, or for an autonomous system to compromise our critical infrastructure, before we decide that emergency powers were necessary after all.
Preparedness is not alarmism.
Preparedness is governance.
This amendment is a proportionate, necessary, and common‑sense measure to protect the British public and safeguard our national security in an era of rapidly advancing AI capabilities.
I urge Honourable Members across the House to support it.
Britain must not be left powerless in an AI emergency.
This amendment ensures that we are not.
Many devices need the capacity to run without cloud computing. For example, nuclear power plants, hydroelectric, dams, and even your homes heating system need to be able to operate without cloud computing. That is to say we need to be able to shut down data centers and cloud computer without the risk of creating chaos.