Sanders introduces bill to permanently ban superintelligence as Amodei, Altman, and Musk argue for slower AI development
Senator Bernie Sanders announced legislation on September 15, 2026 to permanently ban superintelligence, framing the proposal as a response to what he described as an unregulated race among frontier AI developers. The bill was unveiled at the Future of Life Institute’s Pro-Human Assembly and marks the first formal legislative push in either chamber of Congress aimed at prohibiting the development of superintelligence.
Amodei’s ‘We Must Pace the Frontier’ essay sets the tone
On September 12, Anthropic CEO Dario Amodei published an essay titled ‘We Must Pace the Frontier,’ arguing that frontier AI development must be deliberately slowed so that alignment, security, and evaluation work can catch up. Amodei warned that capability advances are outpacing the field’s ability to measure and mitigate emerging risks, and called for binding procedural constraints rather than voluntary commitments. The essay was the clearest public statement from a frontier-lab chief executive that technical readiness, not commercial release calendars, should determine deployment timing. Within 48 hours, the position drew formal support from at least one other major industry figure and the first joint regulatory posture ever adopted by Anthropic, Microsoft, and OpenAI together.
Nadella and Altman add public support
By Monday September 14, Microsoft CEO Satya Nadella had written publicly that Microsoft welcomes the deliberate pacing Amodei proposed, framing the slowdown as compatible with Microsoft’s enterprise roadmap and cloud commitments. Earlier, on September 12, OpenAI CEO Sam Altman told Fortune that OpenAI will not go public in 2026, citing ‘given everything happening with safety’ as the reason for delaying the offering. The two statements, read alongside Amodei’s essay, mark the first time Anthropic, Microsoft, and OpenAI have jointly endorsed mandatory third-party audits with a government pause power — a regulatory posture none of the three firms held as recently as July 2026. Elon Musk, who has long advocated slower frontier development, has also publicly argued for a more cautious timeline, aligning his position with the other frontier-lab leaders.
Trump weighs in from the White House
On September 14, President Trump posted on Truth Social calling Amodei ‘a perfect little angel’ and asserting that his administration has ‘tremendous criminal and regulatory power over AI firms.’ The post, which did not reference the proposed legislation directly, signalled openness to using existing federal authorities against AI companies whose products or practices are deemed unsafe. White House staff did not clarify whether the statement constituted an endorsement of the Sanders bill or a separate warning to frontier developers. The intervention adds a federal-executive dimension to a debate that until this month had been driven primarily by company executives and members of Congress.
Industry consensus on audits and pause power
According to statements and essays published between September 12 and September 15, Anthropic, Microsoft, and OpenAI now jointly hold the first consensus among frontier labs on mandatory third-party audits accompanied by a government pause power. The three companies previously diverged on the scope of pre-deployment evaluation, with OpenAI historically favouring internal red-teaming and Anthropic supporting independent review. The new shared posture is narrower than Sanders’ legislation but represents the first time the three largest U.S. frontier developers have aligned on an enforceable compliance structure. Proponents describe the framework as a procedural floor; critics argue it leaves the definition of frontier systems to the firms being regulated.
Sanders’ superintelligence ban bill and the legislative path
Sanders’ bill, announced at the Future of Life Institute’s Pro-Human Assembly on September 15, would prohibit the development of superintelligence as defined by capability thresholds rather than by company or model name. The proposal goes further than the audit-and-pause framework now backed by Anthropic, Microsoft, and OpenAI, and would impose a categorical prohibition rather than a procedural slowdown. Speaker of the House Mike Johnson signalled on September 15 that the House will consider a narrow data-centre bill next week and will not take up a broader framework in the near term. The chamber-by-chamber divergence means the bill faces an uphill path even as industry support for slower development has broadened.
Compute scale underscores what is at stake
Anthropic recently disclosed $517 billion in compute commitments and 14.8 gigawatts of forward capacity ahead of its anticipated IPO, figures that quantify the infrastructure now tied to the pacing debate. The disclosures, made in regulatory filings in the weeks before the September announcements, indicate that frontier AI development is backed by physical and financial commitments measured in years rather than quarters. Industry executives have pointed to that scale in arguing that any pause mechanism must be calibrated to avoid stranding contracted capacity, while safety advocates have cited the same figures as evidence that binding limits are overdue. The combined weight of disclosed compute commitments, executive statements, and the new Sanders bill has reframed the pacing fight from a technical argument into a concrete legislative and capital-allocation question. With Anthropic, Microsoft, and OpenAI now aligned on enforceable audits and a government pause power, and with Sanders formally introducing a superintelligence ban, the September 2026 announcements have produced the first shared regulatory vocabulary among frontier developers and the first congressional proposal to permanently ban superintelligence.