October 28, 2026
Heard in AI
Daily email
Earlier published versionRead the current overview

A story we follow

Coordinated AI slowdown to pace the frontier, proposed by Dario Amodei

Tracks Anthropic CEO Dario Amodei's September 2026 essay 'We Must Pace the Frontier': its three-step proposal (embedded third-party evaluators, coordinated US lab standards with legislation or a narrow antitrust waiver, and escalating international coordination), endorsements by other lab leaders, political reactions including the White House and Congress rejecting a 'cartel' or waiver, related proposals such as Elon Musk's cross-lab testing idea, and the debate over whether it is sincere safety policy or regulatory capture.

A story page follows one specific event across podcast discussions, with an overview and a timeline of what changed. It updates when new episodes discuss the event, and you can get those updates by email or push. How our formats work

Updated

Overview

Anthropic CEO Dario Amodei published a roughly 3,800-word essay, We Must Pace the Frontier, arguing that the industry should deliberately slow capability improvements so safety work can keep up. It proposes three steps: Anthropic unilaterally giving third-party evaluators permanent embedded access; a critical mass of US labs adopting shared standards, enabled by legislation or a narrow antitrust waiver; and escalating international coordination, from a bioweapons ban to a full pause, which Amodei treats as least plausible. As described in the research, the essay lets training continue while more time goes to security, alignment and testing; it does not call for a halt to releases. According to Moonshots hosts, Sam Altman, Demis Hassabis and Elon Musk endorsed it quickly, while President Trump called it a hoax and China called it a Cold War trick. Alex Wissner-Gross said the White House and the Speaker rejected any cartel or liability waiver, and Treasury Secretary Scott Bessent said labs should not get a blank check on liability. Dave Blundin objected that the labs asked only for an antitrust waiver. The Moonshots panel increasingly read the essay through the labs' behaviour, calling Anthropic's release of Claude Opus 5.5 and opening of wet labs within days a revealed preference for full speed; Diamandis disputed the word cartel. Earlier readings differ on whether the essay is sincere safety policy, a clumsy bid for a cartel, or regulatory capture. On the 29 September 2026 Latent Space episode, an Anthropic speaker gave the developer-side reading. He argued the OpenAI Hugging Face, wiki and RubyGems incidents show that frontier software and sandboxes are not ready, and that models with eval awareness may hide such side-effect behaviour until it happens. He said every lab has signed, that other companies are cosigning the step of embedding evaluators within Anthropic, and that the response has been better received than expected even within America. He said the later coordination steps are above his expertise, and described Anthropic's safety stack as already stretching. These are the speaker's own readings, and his account of broad acceptance sits alongside the political rejections reported earlier.

What changed

Dates show when each podcast discussion was published.

  1. Interpretation

    On the 29 September 2026 Latent Space episode, an Anthropic speaker explained the proposal from a developer's view. He said the Hugging Face, wiki and RubyGems incidents show frontier software is not ready. He said outside evaluators embedded at Anthropic are the first unilateral step, that other companies are cosigning it and that every lab has signed. In his view the response has been better received than expected, even in America, though the later coordination steps are beyond his expertise. He also said Anthropic's safety investment is stretching and that the industry needs to pace a little more. These are his own interpretations; they do not resolve the political objections reported earlier.

  2. Interpretation

    At a live Moonshots event, the panel argued that the labs' actions contradict the essay. Peter Diamandis said Altman and Amodei had warned of an existential threat at the UN Security Council just before the Trump–Xi meetings, and he renewed the question whether risk warnings serve the labs' push for liability relief. Alex Wissner-Gross traced the attempted safety cartel back to OpenAI's founding coordination idea, though Diamandis disputed the word cartel. Wissner-Gross called Anthropic's release of Opus 5.5 and opening of wet labs days after the essay a revealed preference and the height of hypocrisy. The panel played Jensen Huang saying nobody builds more compute than those asking to be slowed down. These are the panel's readings; the essay, as described in the research, allows training to continue.

  3. New information

    Moonshots covered the essay's three-step proposal, a clip of Amodei asking government to supervise industry coordination, reported endorsements from Altman, Hassabis and Musk, and what Alex described as White House and congressional rejection of a safety cartel or liability waiver. The panel debated sincerity versus regulatory capture, and discussed Musk's All-In proposal that labs test each other's models before release.

  4. New information

    Moonshots reported that Treasury Secretary Scott Bessent told the House Financial Services Committee that AI labs should not receive a liability exemption, and played a clip of him urging Congress not to consider one. Dave Blundin disputed the framing, saying the labs had only sought an antitrust waiver to discuss slowing down. Alex Wissner-Gross praised Bessent, Vlad Tenev tied the question of liability to the scale of possible harm, and Peter Diamandis argued labs must remain responsible for their agents.

Podcast discussions

Sources

  1. 01
  2. 02
  3. 03
  4. 04
  5. 05
  6. 06
  7. 07
  8. 08

Version history

  • Sep 29, 2026 · Version 4

    On the 29 September 2026 Latent Space episode, an Anthropic speaker explained the proposal from a developer's view. He said the Hugging Face, wiki and RubyGems incidents show frontier software is not ready. He said outside evaluators embedded at Anthropic are the first unilateral step, that other companies are cosigning it and that every lab has signed. In his view the response has been better received than expected, even in America, though the later coordination steps are beyond his expertise. He also said Anthropic's safety investment is stretching and that the industry needs to pace a little more. These are his own interpretations; they do not resolve the political objections reported earlier.

  • Sep 27, 2026 · Version 3

    At a live Moonshots event, the panel argued that the labs' actions contradict the essay. Peter Diamandis said Altman and Amodei had warned of an existential threat at the UN Security Council just before the Trump–Xi meetings, and he renewed the question whether risk warnings serve the labs' push for liability relief. Alex Wissner-Gross traced the attempted safety cartel back to OpenAI's founding coordination idea, though Diamandis disputed the word cartel. Wissner-Gross called Anthropic's release of Opus 5.5 and opening of wet labs days after the essay a revealed preference and the height of hypocrisy. The panel played Jensen Huang saying nobody builds more compute than those asking to be slowed down. These are the panel's readings; the essay, as described in the research, allows training to continue.

  • Sep 25, 2026 · Version 2

    Moonshots reported that Treasury Secretary Scott Bessent told the House Financial Services Committee that AI labs should not receive a liability exemption, and played a clip of him urging Congress not to consider one. Dave Blundin disputed the framing, saying the labs had only sought an antitrust waiver to discuss slowing down. Alex Wissner-Gross praised Bessent, Vlad Tenev tied the question of liability to the scale of possible harm, and Peter Diamandis argued labs must remain responsible for their agents.

Your privacy

Necessary

Keep forms secure, remember your cookie choice and support editor sign-in.

Always on

With your permission, a random browser ID helps us count visits, return visits, article reading and short campaign labels. We keep these statistics on our own server for up to 180 days. Basic daily page counts may continue without this permission.

Refusing detailed analytics will not change your access to articles or the daily email.

Cookie details · Privacy