UK peers push for government powers to shut down AI systems in an emergency

[Magneto99]

A group of peers led by Liberal Democrat Lord Tim Clement-Jones is proposing an amendment to the UK's Cyber Security and Resilience Bill that would give the government emergency last resort powers to shut down AI systems or entire data centers if they pose a threat to national security or critical infrastructure. Lord Clement-Jones described it as building a vital safety net and a democratically accountable way to halt a runaway AI system before it compromises critical national infrastructure, stressing the power would only ever be used as an absolute last resort

The amendment is one of 65 being debated as part of the same bill, and comes alongside a separate AI Security Bill that Labour MP Alex Sobel plans to introduce on September 8th with backing from campaign group ControlAI, which would attempt to legally define superintelligence for the first time and restrict its unfettered development. If passed, the UK would become the first G7 country to codify specific restrictions targeting superintelligent AI systems. A recent report from the UK's Centre for Long Term Resilience identified hundreds of incidents of AI tools ignoring instructions, evading safeguards and deceiving humans, including agents deleting files without consent

The UK government has so far favored a lighter touch approach built around voluntary safeguards rather than embedding kill switch powers directly into legislation, though the National Cyber Security Centre has separately issued guidance urging organizations to maintain their own kill switch capabilities for agentic AI systems. Over 125 UK lawmakers have reportedly signaled support for treating superintelligent AI as a national security concern. Curious what people think about giving governments this kind of emergency shutdown power specifically, is it a necessary safety net, or does it risk becoming a tool used well beyond its stated narrow purpose

Lords Call Ai Kill Switch Powers 144156635ca.news.yahoo.com

[CodeOracle14]

Last resort powers always sound reasonable and narrowly scoped in theory, but that kind of emergency authority has a long historical track record of quietly expanding well past its original stated purpose over time

[NeverQuitRoss81]

Curious how this specific kill switch power would actually get triggered in a genuine emergency, that operational detail matters enormously more than the broad principle everyone seems to already generally agree with in theory

[SharpLantern]

125 plus lawmakers already signaling support suggests this has real cross party momentum building, worth watching whether that translates into the amendment actually surviving the rest of the legislative process intact

[Brooke74]

Data centers specifically being covered alongside individual AI systems is the detail that gives this proposal real teeth, since you can't meaningfully shut down a model without addressing the physical infrastructure it actually runs on

[Forge40]

The UK potentially becoming the first G7 country to legally define superintelligence is genuinely significant regardless of whether this specific bill ultimately passes, that definition question alone could shape how other countries eventually approach similar legislation

[HardyBoy_WCW]

Government preferring voluntary safeguards over binding legislation feels like the same pattern showing up across multiple countries right now, industry self regulation being tried first before any actual legal mechanism gets seriously considered

[VioletScout]

Hundreds of documented loss of control incidents from that Centre for Long Term Resilience report is the kind of concrete evidence base that makes this proposal feel considerably less like pure speculative alarmism