- Domestic kill switches are no longer feasible, UK government warns
- AI companies and government need to work together to reduce the risks
- OpenAI and 100+ other companies warn of “limited window to strengthen” defenses
The UK Government has reportedly rejected calls to create an emergency AI kill switch, per BBC reporting, and it’s largely because there are simply too many parties and factors to consider at this stage.
The Cabinet Office told the BBC that it “cannot simply turn AI off” because preventing access to a model or shutting down certain infrastructure would not stop the same model being developed and hosted, thus misused, elsewhere.
This is in response to MPs pushing for a statutory trigger that the government could pull in the event of an emergency.
UK government rules out AI kill switch
The government also implied that the responsibility lies with AI companies, and that developers have their own responsibility to build products with the appropriate safeguards.
Former OpenAI researcher Danial Kokotajlo also agrees that one country building a legal or technical kill switch would not be feasible, but Kokotajlo does see companies working together as one potential solution.
“Switching off access to an AI model in an emergency will do little to protect you,” Kokotajlo explained in relation to a domestic kill switch. “You’re still going to be steamrolled by the super intelligences created in the US.”
The latest kill switch debates have emerged from a September 1 discussion in the House of Lords, partly fueled by an open letter warning of a “limited window to strengthen cyber defenses.” More than 100 companies, including OpenAI, Anthropic, Google, Microsoft, AWS and Adobe issued the stark warning, urging governments, cybersecurity companies and frontier AI developers to come together.
“Our security services and regulators possess no specific agile statutory mechanism to compel a physical or digital shutdown,” Lord Clement-Jones warned earlier this month about the potential of a rogue AI.
As for what’s next, while an AI kill switch may be technically infeasible, we can certainly hope that the companies and bodies at the forefront of AI and legislation collaborate more effectively to reduce future risks.
