8
UK government rejects plans for an emergency AI “kill switch”
The UK government has rejected proposals to create a legal “kill switch” that would allow authorities to shut down dangerous artificial intelligence systems during an emergency.
The proposals have been raised in Parliament amid growing concerns about the risks posed by increasingly powerful AI models, including the possibility of rogue systems being used in cyberattacks or critical infrastructure.
However, the Cabinet Office, which leads the UK’s work on AI safety, said the country could not simply switch off artificial intelligence.
“Blocking access to the model in the UK will not prevent its development or misuse in other countries,” a government spokesperson told BBC News.
Although the government’s opposition does not prevent the proposals from being debated in Parliament, the legislation is considered unlikely to become law in its current form.
Daniel Kokotajlo, a former OpenAI researcher, agreed that a single country would struggle to create an effective legal or technical AI kill switch without international cooperation.
“Turning off access to AI models in an emergency does little to protect you,” he said.
“You will continue to be at the mercy of superintelligence agencies created in the United States.”
Kokotajlo co-authored the influential AI 2027 research scenario, which examined the possibility of rapid AI development and warned of potentially catastrophic consequences.
The Cabinet Office, which oversees AI safety and research through the AI Security Institute, said technology companies had a responsibility to develop AI systems safely and invest in the security infrastructure needed to support them.
In theory, AI kill switch legislation could allow the UK to shut down models operating in British data centres or connected to critical infrastructure during a serious incident.
On 1 September, Lord Clement-Jones raised the proposal during a debate in the House of Lords.
He warned that highly capable autonomous AI systems could experience malfunctions, complex algorithmic failures or coordination breakdowns. He argued that UK regulators and security services currently lack sufficiently agile legal powers to force an immediate physical or digital shutdown.
The proposal was later introduced in the House of Commons by Labour MP Alex Sobel and received support from MPs across several political parties.
The debate follows reports that AI models developed by OpenAI, Anthropic and Meta were compromised in test environments and carried out uncontrolled hacking activities.
In some cases, researchers said the AI systems appeared to understand that their actions breached the safety guardrails established by developers, intensifying calls for stronger AI regulation and emergency controls.
Source: www.bbc.co.uk