Jack Clark calls for mandatory AI kill switches

Artificial intelligence systems are advancing so rapidly that industry leaders are confronting a fundamental dilemma over how to keep the technology under human control, culminating in fresh calls for mandatory, third-party-verifiable emergency shutdown mechanisms.

The push for government-enforced safeguards follows a series of high-profile departures and internal alarms at major artificial intelligence laboratories. According to reporting from the Bbc, Jack Clark, one of seven co-founders at Anthropic, argued that society should eventually pass rules requiring companies to install a reliable way to pull the plug on dangerous software. Anthropic was formed in 2021 by a group of former employees of its rival OpenAI, and has created the popular chatbot Claude while releasing increasingly capable models.

Media additions

While most major laboratories currently maintain internal methods for shutting down models, Clark emphasized that verification by an independent outside party should form a core component of the broader regulatory debate. The discussions come as the industry faces intense scrutiny over the potential existential threats posed by autonomous agents. A recent resignation by an Anthropic researcher who cited fears of human extinction amplified public anxiety, prompting alignment lead Evan Hubinger to publicly estimate the probability of human extinction from advanced systems within the decade. Jacob Coxon resigned from Anthropic on September 9, 2026, and four days later appeared on television to warn that current off-switches might become obsolete if future autonomous systems operate in a distributed, swarm-like fashion and route around centralized shutdown commands.

Those survival estimates have drawn mixed reactions across the technology sector. Nobel laureate Geoffrey Hinton characterized the probability as reasonable, noting that rogue systems might seize control of critical internet-connected infrastructure. Conversely, other figures have pushed back against the rhetoric. Clement Delangue, leader of developer platform Hugging Face, likened apocalyptic predictions to asking an air conditioning technician about climate change, while Grindr leader George Arison suggested that public alarm is being leveraged by corporations to justify immense market valuations and business plans.

The regulatory response to these developments has exposed deep divisions between political leaders and the technology sector. In the United States, lawmakers have introduced proposed legislation dubbed the Kill Switch Act, which would compel companies to equip tools with emergency shutdown mechanisms and grant government agencies the authority to deactivate non-compliant software. Yet, US President Donald Trump rejected safety warnings during social media statements, dismissing the prospect of technological domination as a hoax and comparing alarms to past political rhetoric surrounding climate concerns. On Monday, 14 September 2026, Trump attacked calls for guardrails on Truth Social and phoned Nvidia chief executive Jensen Huang on speakerphone during the All-In Summit in Los Angeles to reiterate his position. Meanwhile, House Speaker Mike Johnson noted that the president would meet with industry leaders to discuss whether the government has a role to play in ensuring safety.

Meanwhile, the international regulatory landscape remains fragmented. The UK government rejected proposals to mandate a domestic kill switch, arguing that local restrictions would fail to prevent misuse or development elsewhere in the world. Confronted with regulatory resistance, major laboratories have instead pursued self-governance initiatives. Microsoft published an extensive code of conduct detailing operational rules for its in-house models, while Anthropic, Google, and OpenAI have reportedly discussed establishing an independent standards body modeled on financial industry regulators to evaluate frontier models prior to commercial release.

As financial markets react to potential shifts in development velocity, investors and industry observers are weighing what happens next. Market analysts point out that a slowdown in frontier model deployment could alter capital expenditure on data infrastructure and semiconductor manufacturing. Policy discussions are expected to continue as lawmakers review proposed statutory frameworks, and major technology firms work to finalize internal compliance codes ahead of upcoming corporate reporting cycles.