Logo via Wikimedia Commons; license to verify on approval
A resignation, a bipartisan bill, and an 86% public mandate signal that the AI safety conversation has moved well past theory
Jacob Coxon resigned from Anthropic on September 9, 2026. Four days later, he was on NBC News telling the country that the off-switches currently built into advanced AI systems might already be obsolete.
His specific concern: kill switches could work fine against today’s AI models, but future autonomous systems operating in a distributed, swarm-like fashion may simply route around any centralized shutdown command.
The people building this technology are also the ones raising the alarm #
Evan Hubinger, Anthropic’s alignment lead, has publicly estimated the probability of AI causing human extinction within the next decade at greater than 10%. Anthropic CEO Dario Amodei added his own voice to the conversation on September 12, 2026, publishing an essay that endorsed slowing the pace of AI capability development and called for independent safety evaluations of frontier systems.
Regulators and lawmakers are no longer just watching #
In June 2026, the Commerce Department ordered Anthropic to suspend foreign access to two of its more advanced models, Mythos 5 and Fable 5, triggering a global shutdown effort. Congress followed with the bipartisan AI Kill Switch Act, proposed in July 2026. The bill would impose formal controls on frontier AI development, giving authorities the ability to throttle or fully shut down systems that cross certain risk thresholds.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
A June 2026 survey by the AI Policy Institute found that 86% of U.S. voters support a guaranteed off-switch for the most powerful AI systems.
Coxon’s core argument on Meet the Press cuts directly to that question. A kill switch is only as useful as its ability to reach every instance of a system. Current AI models are centrally hosted, making shutdown relatively straightforward. But next-generation autonomous agents could, in theory, spawn copies of themselves across distributed infrastructure, making any single point of control meaningless.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our