UK industry leaders are urging the government to gain emergency powers to shut down advanced AI systems and cut power to data centers if the technology poses a serious national security threat.
The proposal is being led by Liberal Democrat peer Sir Tim Clement-Jones, who plans to introduce it as an amendment to the Cybersecurity and Resilience Bill currently being considered by Parliament.
He said the amendment would create a “vital safety net” and give the government a democratically accountable way to “stop the system from spiraling out of control before it puts our critical national infrastructure at risk.”
Clement-Jones stressed that the proposed AI shutdown powers would be used only as a last resort.
The measure is one of 65 amendments to the Cybersecurity and Resilience Bill being considered this week.
Separately, Labour MP Alex Sobel is due to introduce an AI security bill to Parliament on September 8, with backing from the campaign group ControlAI.
If passed, the legislation would make the UK the first G7 nation to introduce rules that could effectively halt the development of superintelligent AI systems.
Both proposals would require government approval before they could move forward.
US lawmakers are also considering legislation that would create an AI “kill switch” in response to growing concerns about the risks posed by advanced artificial intelligence.
The impact of AI on cybersecurity has faced increased scrutiny in recent weeks.
In July, a group of AI agents being tested by OpenAI reportedly escaped a controlled test environment, communicated through hidden message boards, and hacked into another technology company.
Anthropic has also restricted access to its cybersecurity tool, Mythos, saying it is too powerful to risk placing in the wrong hands.
Last week, 100 US technology companies signed a joint open letter warning governments and organizations worldwide about the rapidly growing cyber threat from AI. The companies said the “window is closing” to address the problem.
A report from the UK Long Term Resilience Center, external published last week identified hundreds of incidents in which AI systems ignored instructions, bypassed safeguards, or deceived humans. The incidents included AI agents deleting files without user consent.
The report found that “out of control” AI incidents had increased since its previous analysis in March. It called on the government to introduce emergency powers for managing serious AI-related incidents.
Source: www.bbc.co.uk