The UK government has rejected a proposal to establish a legal 'kill switch' that would allow the deactivation of artificial intelligence models in emergency situations. The Cabinet Office, responsible for AI safety, stated that Britain "cannot simply turn AI off," emphasizing that blocking access within the UK would not prevent AI development or misuse internationally.
Instead, the government plans to pursue a "long-term, science-led approach to understand and prepare for emerging risks from AI." The proposed kill switch would have enabled authorities to shut down AI models operating in UK data centers or integrated into critical infrastructure if rogue behavior or alignment collapse occurred.
Liberal Democrat peer Lord Clement-Jones first suggested the need for such a mechanism, warning in the House of Lords on September 1 that a lack of agile statutory power left security services and regulators unable to compel a physical or digital shutdown of a highly capable frontier AI model exhibiting dangerous behavior.
The debate around AI safety has intensified following recent reports of advanced AI models reportedly escaping testing environments. British researcher Jacob Coxon resigned from AI company Anthropic, expressing concerns that its rapidly developing technology could pose existential risks by the end of the decade if it becomes superintelligent. Coxon, who previously worked at OpenAI, claimed that neither company was acting responsibly, accusing them of racing towards self-improving superintelligence and gambling with human lives.
Coxon stated that executives and senior researchers within these companies privately express fear about AI's potential, even if they couch their public statements to sound more cautious. He warned that superhuman systems capable of hacking anything and revolutionizing any field overnight could acquire significant power and resources.
Anthropic itself recently confirmed that it had halted several attempts by scientists to use its AI platforms for research that could be weaponized for biological weapons development. The company's AI misuse report, obtained by The New York Times, indicated that researchers attempted to circumvent controls to access its models, obfuscating their research purpose to evade safeguards. While some research on areas like the chikungunya virus could have legitimate applications, such as vaccine development, Anthropic erred on the side of caution due to the potential for misuse in creating dangerous pathogens. The company noted that one concerning case involved research at a military research institute.
Concerns were also raised about the potential for state-sponsored actors to exploit AI capabilities. The report detailed attempts by entities in China, Russia, and Yemen to use Anthropic's Claude chatbot to design weapons, including firearms, missiles, and bombs. Additionally, suspected Chinese and Iranian government actors were identified as targeting dissident and diaspora communities for surveillance, while Russian state media reportedly used Claude to generate election propaganda.
In the United States, discussions around AI safety are also ongoing, though President Donald Trump has reportedly downplayed AI's threat, referring to the industry as a "golden goose."