Jacob Coxon’s departure from Anthropic on September 9, 2026, has directly triggered a fresh legislative push in the US Senate to ban the development of superintelligence. The move follows Coxon's public resignation, in which he claimed that both Anthropic and OpenAI are prioritizing rapid scaling over the safety of human lives. In response, lawmakers are drafting a bill that would move beyond simple oversight, seeking to outlaw the creation of AI models that exceed specific computational and cognitive thresholds considered to be 'superintelligent.'
The resignation of a high-level researcher like Coxon has shifted the debate in Washington from voluntary safety agreements to mandatory federal prohibitions. Coxon’s allegations suggest that internal safeguards at top AI firms are insufficient to mitigate existential risks, providing the political ammunition necessary for senators to argue that the industry is incapable of self-regulation. This specific bill represents the most aggressive stance taken by US regulators to date, targeting the hardware and training runs required for next-generation AI.
From a geopolitical perspective, the bill is controversial, as critics argue a domestic ban could allow international rivals to take the lead in the AI arms race. However, the Senate's focus remains on the immediate risks highlighted by Coxon’s exit. The proposed legislation focuses heavily on 'superintelligence' as a unique category of risk, distinguishing it from general-purpose AI used in current commercial applications. This distinction is intended to protect existing tech industry growth while stopping the most dangerous experimental developments.
For the cryptocurrency and blockchain sectors, this regulatory pivot carries significant weight. Many decentralized compute projects (DePIN) and AI-integrated protocols operate at the intersection of high-performance computing and machine learning. While the bill primarily targets centralized labs like Anthropic, the broader regulatory chilling effect could impact how decentralized GPU networks are utilized. If 'superintelligence' training is banned on a federal level, the demand for permissionless, decentralized compute resources could either surge or face its own set of strict compliance hurdles.
Investors and developers should watch the Senate Judiciary Committee for the official introduction of the bill's text. The most critical factor will be the technical definition of 'superintelligence' used by lawmakers, as this will determine which training architectures are prohibited. Furthermore, testimony from other industry whistleblowers or tech CEOs in the coming weeks will likely decide if this bill reaches a floor vote before the end of the 2026 legislative session.