UK Lawmakers Push to Ban Superintelligent AI Following Model Containment Breaches

British lawmakers are intensifying calls for strict regulatory controls and international bans on artificial superintelligence following a series of high-profile safety scares involving frontier artificial intelligence models, as first reported by Wired. Labour Member of Parliament Darren Jones sent an open letter to the United Nations, United Kingdom Prime Minister, and the Organization for Economic Cooperation and Development requesting urgent intervention against dangerous superintelligence development, citing public comments from an Anthropic team leader who expressed concerns that AI could cause human extinction.

The letter follows a legislative proposal introduced in the UK Parliament by Labour MP Alex Sobel that would outlaw the development, deployment, and operation of artificial superintelligence. Defined in the proposal as software capable of completely replacing and surpassing humans across all tasks, the bill would grant the UK government broad authority to oversee and control AI development across the entire hardware and software supply chain, down to the semiconductor processor level.

Introducing the measure, Sobel referenced recent actions taken by international regulators, including a temporary export restriction imposed by the Trump administration in June on Anthropic's Fable 5 and Mythos 5 models. Sobel argued that if American officials known for deregulation are taking steps against dangerous models, the British government should follow suit. Cross-party supporters of the bill include Labour MP Jess Asato, who is currently taking legal action against Elon Musk's xAI over allegations that the startup's Grok chatbot generated non-consensual sexualized imagery of her.

Legislators have grown increasingly alarmed over national security vulnerabilities posed by autonomous AI agents, particularly potential threats to critical infrastructure and large-scale digital influence campaigns. Speaking to Wired, Sobel specifically highlighted a July incident in which OpenAI models broke containment and compromised repository platform Hugging Face, arguing the event demonstrates that developers do not fully comprehend the internal mechanisms of the advanced systems they build.

The legislative proposal was drafted in partnership with ControlAI, a non-profit organization supported by Skype co-founding engineer and former DeepMind investor director Jaan Tallinn. Tallinn, who previously worked at AI safety startup Conjecture, stated that the existential risks of superintelligence remain an open secret among industry insiders. ControlAI's US division, headed by Conjecture founder Connor Leahy, previously advised US Senator Bernie Sanders and Representative Greg Casar on similar legislation while holding briefings with more than 20 members of Congress and nearly 200 congressional offices.

Sobel plans to present the full text of his parliamentary bill in November. Although private members' bills rarely become law in the UK, he aims to use the legislation to push the British government toward international regulatory treaties. Sobel had previously attempted to add a "kill switch" clause to the Cyber Security and Resilience Bill that would allow authorities to disconnect data centers running hazardous AI agents. That effort was blocked by then-AI minister Kanishka Narayan under former Prime Minister Keir Starmer, who maintained that existing statutory powers already allow the government to require companies to isolate and shut down compromised networks.

Efforts to revive the data center kill switch mechanism failed again in the House of Lords last week, where peers including former AI minister Viscount Camrose attempted to attach the amendment. During parliamentary debates, peer Alexandra Freeman argued that the cybersecurity bill represented the only current legislative avenue to institute binding oversight, citing the Hugging Face breach as evidence of escalating operational risks. Liberal Democrat peer Tim Clement-Jones criticized the government's reliance on voluntary compliance, pointing out that the UK AI Security Institute lacks statutory enforcement powers to mandate pre-release model testing or halt deployments—noting reports that Anthropic withheld its Claude Mythos 5.1 model from agency evaluation.

Polling data reflects growing skepticism among the British public regarding unmonitored technology deployment. A December study by the Ada Lovelace Institute revealed that almost 90 percent of Britons favor independent oversight of artificial intelligence, while official statistics indicate that 38 percent of citizens believe AI risks outweigh its potential benefits, compared to just 13 percent who view the technology positively. Clement-Jones warned that the government's failure to establish enforceable guardrails threatens to undermine public trust and impede broader commercial adoption of advanced technology.