Jacob Coxon, a researcher at Anthropic, resigned on September 9, 2026. He stated that existing kill switches for AI may become ineffective as future autonomous systems could operate in distributed, swarm-like architectures that bypass centralized shutdown commands.
Evan Hubinger, the alignment lead at Anthropic, estimated the risk of AI-induced human extinction within the next decade at over 10%. On September 12, 2026, CEO Dario Amodei published an essay advocating for slower development of AI capabilities and independent safety assessments.
In June 2026, the U.S. Commerce Department mandated that Anthropic restrict foreign access to its Mythos 5 and Fable 5 models. Subsequently, in July 2026, Congress introduced the bipartisan AI Kill Switch Act to regulate frontier AI development.
The proposed legislation aims to grant authorities the power to throttle or deactivate systems that exceed specific risk levels. A survey conducted by the AI Policy Institute in June 2026 indicated that 86% of U.S. voters support mandatory off-switches for powerful AI.
