Fazlur Rahman
Published September 13, 2026 - Washington, DC.
Bill text and Sanders quotes; Amodei essay from darioamodei.com; Pachocki voluntary-slowdown remarks September 6; UK open letter details; data on board and compute-monitoring regime last verified September 2026.
Senator Bernie Sanders of Vermont introduced a bill in September 2026 that would ban the development of artificial superintelligence and impose a pause on training advanced AI systems above a defined capability threshold. The bill follows a September 12 essay by Anthropic CEO Dario Amodei titled We Must Pace the Frontier, which was publicly backed by OpenAI's Sam Altman, xAI's Elon Musk, and Google DeepMind's Demis Hassabis. The bill is also backed by OpenAI chief scientist Jakub Pachocki and by an open letter to UK mayor Andy Burnham calling for a multinational treaty on superintelligence. Most Washington observers consider the bill unlikely to pass in 2026, but it is widely seen as the marker for a much larger policy debate.
What is in the bill
The bill would create a federal AI Safety Board with the power to define capability thresholds, mandate pre-training evaluations, and block the deployment of any system that exceeds the threshold. Companies would be required to publish detailed safety reports, share evaluation data with the board, and accept on-site inspections. The bill would also establish a federal compute-monitoring regime, similar to the export-control regime for advanced semiconductors, to track large training runs. Civil penalties for non-compliance would start at 1 percent of global revenue and scale to criminal liability for executives (Reuters, Politico, September 2026).
Sanders introduced the bill following a series of high-profile resignations from frontier AI labs, public warnings from researchers about the risks of self-improving superintelligence, and the publication of Dario Amodei's essay. In a BBC Newsnight interview, Sanders said: you have got to be a moron not to say, slow it down, reflecting the view that AI is the most important labour and economic issue facing the country and that the United States is rushing into a technology that could destroy millions of jobs (BBC Newsnight, September 2026).
What Amodei proposed
On September 12, 2026, Anthropic CEO Dario Amodei published a 13,000-word essay titled We Must Pace the Frontier on his personal website. The essay argues that AI capability is improving faster than safety, and proposes a three-step plan. The first step is embedded evaluators, third-party organisations like the Alignment Research Center's METR that have employee-like access to labs and can flag risks before deployment. The second step is industry safety coordination under antitrust exemptions, so that labs can share safety data without violating competition law. The third step is global coordination on a relative-speed-limit treaty similar to nuclear arms-control regimes like the SALT treaties (darioamodei.com, The Verge, September 2026).
Amodei warned that a misaligned AI swarm could take over the entire internet in 6-12 months and cause hundreds of billions of dollars in damage, and that the geopolitical case for slowing down is also a case for keeping the United States ahead of the Chinese Communist Party. The essay was publicly backed on September 13 by Sam Altman of OpenAI, who said we will do the same; by Elon Musk of xAI, who called the essay the most important paper of the year; and by Demis Hassabis of Google DeepMind, who said the proposal was thoughtful and overdue (darioamodei.com, Forbes, September 2026).
Industry and political backing
| Actor | Position | Date |
|---|---|---|
| Sen. Bernie Sanders (US) | Bill to ban ASI, pause advanced training | Sept 2026 |
| Dario Amodei (Anthropic) | Essay calling for slowdown | Sept 12, 2026 |
| Sam Altman (OpenAI) | Backed Amodei plan; we will do the same | Sept 13, 2026 |
| Elon Musk (xAI) | Backed Amodei plan; most important paper of the year | Sept 13, 2026 |
| Demis Hassabis (Google DeepMind) | Backed Amodei plan; thoughtful and overdue | Sept 13, 2026 |
| Jakub Pachocki (OpenAI) | Called for voluntary slowdowns | Sept 6, 2026 |
| UK open letter to Andy Burnham | Multinational treaty on superintelligence | Sept 2026 |
What the industry is saying
OpenAI chief scientist Jakub Pachocki called for voluntary slowdowns on advanced AI training on September 6, 2026, in remarks at a closed-door conference in San Francisco, a notable shift for a company that has historically resisted mandatory pauses. The shift is widely attributed to internal concerns about the speed of capability gains, and to the broader alignment debate that has spilled out of Anthropic following the high-profile resignations of researchers including Jacob Coxon and Evan Hubinger (Forbes, The Verge, September 2026).
In the United Kingdom, an open letter to Greater Manchester mayor Andy Burnham, who is widely expected to become the next Prime Minister, calls for a multinational treaty on superintelligence. The letter is signed by leading UK and European AI researchers, civil-society groups, and several members of the House of Lords. Burnham has not yet publicly responded, although allies say he is open to the idea of an international AI safety treaty modelled on the Nuclear Non-Proliferation Treaty (DW, BBC, September 2026).
What the bill would actually do
If passed, the bill would create the most powerful AI oversight regime in the world. The AI Safety Board would have the power to define what counts as artificial superintelligence, mandate pre-training evaluations, and block the deployment of any system that exceeds the threshold. Companies would be required to publish detailed safety reports, share evaluation data with the board, and accept on-site inspections. The compute-monitoring regime would track large training runs in real time, similar to the export-control regime for advanced semiconductors. Civil penalties for non-compliance would start at 1 percent of global revenue and scale to criminal liability for executives (Reuters, Politico, September 2026).
The bill is modelled in part on the 1958 Defence Production Act, which gives the federal government the power to intervene in critical industries during national emergencies. Supporters argue that AI is now a critical industry, and that the bill is needed to prevent a race to the bottom between labs that prioritises speed over safety. Critics argue that the bill would entrench the market position of the largest labs, chill open-source research, and cede ground to Chinese competitors (Reuters, Politico, September 2026).
What the chances of passage look like
Most Washington observers think the bill as drafted is unlikely to pass in 2026, given the partisan composition of the House and Senate and the strong lobbying presence of frontier AI labs. But the bill is widely seen as a marker for a much larger policy debate that will play out over the next two years, and several provisions, including the federal compute-monitoring regime, are likely to be included in must-pass defence authorisation legislation. Industry executives have signalled they would prefer a federal framework to a patchwork of state laws, which may yet open a path to a compromise bill (Reuters, Politico, September 2026).
What to watch next
Three signals will indicate whether the slowdown movement is gaining real political traction. First, whether the Senate Commerce Committee schedules a markup of the bill before the November midterms. Second, whether the UK open letter to Burnham translates into a formal position from the next British government. Third, whether the frontier labs, including OpenAI, Anthropic, and Google DeepMind, sign on to the embedded-evaluator scheme proposed by Amodei, which would mark the first concrete industry commitment to the slowdown agenda. Until then, the bill is more a marker than a law, but it is the most ambitious marker ever introduced in the United States (BBC, Reuters, Politico, September 2026).
FAQ
What is the Sanders AI slowdown bill 2026?
Senator Bernie Sanders of Vermont introduced a bill in September 2026 that would ban the development of artificial superintelligence and impose a pause on training advanced AI systems above a defined capability threshold. The bill would create a federal oversight board, require independent safety evaluations before any frontier model is trained, and impose civil and criminal penalties on companies that fail to comply. The bill is the most ambitious piece of US AI safety legislation ever introduced in Congress and reflects months of behind-the-scenes work with labour unions, civil-society groups, and a small number of safety-focused researchers (BBC, The Verge, Reuters, September 2026).
Why did Sanders introduce the AI bill?
Sanders has argued for years that AI is the most important labour and economic issue facing the country, and that the United States is rushing into a technology that could destroy millions of jobs and pose an existential risk to humanity. In a BBC Newsnight interview, Sanders said: you have got to be a moron not to say, slow it down. The bill follows a series of high-profile resignations from frontier AI labs, public warnings from researchers about the risks of self-improving superintelligence, and the publication of a major essay by Anthropic CEO Dario Amodei calling for an industry-wide slowdown (BBC Newsnight, Forbes, September 2026).
What does the Amodei essay on slowing AI say?
On September 12, 2026, Anthropic CEO Dario Amodei published a 13,000-word essay titled We Must Pace the Frontier on his personal website. The essay argues that AI capability is improving faster than safety, and proposes a three-step plan: embedded evaluators with employee-like access inside labs, industry safety coordination under antitrust exemptions, and global coordination on a relative-speed-limit treaty similar to nuclear arms-control regimes. Amodei warned that a misaligned AI swarm could take over the entire internet in 6-12 months and cause hundreds of billions of dollars in damage (darioamodei.com, The Verge, September 2026).
Who supports the AI slowdown?
Sanders's bill is backed by OpenAI chief scientist Jakub Pachocki, who on September 6, 2026 called for voluntary slowdowns on advanced AI training. Sam Altman of OpenAI, Elon Musk of xAI, and Demis Hassabis of Google DeepMind have publicly backed Amodei's slowdown plan. In the United Kingdom, an open letter to Greater Manchester mayor Andy Burnham, who is widely expected to become the next Prime Minister, calls for a multinational treaty on superintelligence. The letter is signed by leading UK and European AI researchers, civil-society groups, and several members of the House of Lords (BBC, DW, Forbes, September 2026).
What would the Sanders AI bill actually do?
The bill would create a federal AI Safety Board with the power to define capability thresholds, mandate pre-training evaluations, and block the deployment of any system that exceeds the threshold. Companies would be required to publish detailed safety reports, share evaluation data with the board, and accept on-site inspections. The bill would also establish a federal compute-monitoring regime, similar to the export-control regime for advanced semiconductors, to track large training runs. Civil penalties for non-compliance would start at 1 percent of global revenue and scale to criminal liability for executives (Reuters, Politico, September 2026).
Will the Sanders AI bill pass?
Most Washington observers think the bill as drafted is unlikely to pass in 2026, given the partisan composition of the House and Senate and the strong lobbying presence of frontier AI labs. But the bill is widely seen as a marker for a much larger policy debate that will play out over the next two years, and several provisions, including the federal compute-monitoring regime, are likely to be included in must-pass defence authorisation legislation. Industry executives have signalled they would prefer a federal framework to a patchwork of state laws, which may yet open a path to a compromise bill (Reuters, Politico, September 2026).
