AI chiefs Altman, Amodei, Musk agree to 'pace the frontier'
TL;DR
- OpenAI's Altman, Anthropic's Amodei, xAI's Musk, and DeepMind's Hassabis endorsed Amodei's essay 'We Must Pace the Frontier.'
- The three steps: employee-like access for external evaluators, common safety standards, and international coordination on the rate of progress.
- Critics called the informal deal a 'cartel'; Anthropic's own timeline still puts recursive self-improvement in early 2027.
OpenAI's Sam Altman, Anthropic's Dario Amodei, xAI's Elon Musk and Google DeepMind's Demis Hassabis loosely agreed over the weekend that frontier AI development needs to slow down, all four endorsing Amodei's essay titled 'We Must Pace the Frontier.' The Verge frames the informal pact as either a genuine safety concession or, in critics' words, 'an outright cartel.'
Amodei's proposal is a three-step ask: each frontier lab gives external evaluators 'employee-like access,' the labs agree common safety standards, and any slowdown is coordinated internationally. 'We must slow the pace at which we improve the capabilities of AI models,' Amodei wrote. Altman echoed him on X: 'committing to having independent evaluators with employee-like access is a great idea, and we will do the same.' Musk quote-posted 'Dario is right.' Hassabis called the essay the 'right path forward' at a 'critical moment.'
The pushback is structural. A voluntary agreement among the four best-funded labs is also, mechanically, a way to kneecap the open-source movement and would-be entrants while keeping statutory safeguards at bay. Critics interviewed by The Verge framed it exactly that way.
Nick Reese, formerly at DHS and now at NYU, told the outlet 'the industry as a whole needs new champions.' Apollo Research's Marius Hobbhahn said it is 'one of the best things for safety in a long time if it actually happens.' The conditional there is doing all the work. Daniel Kokotajlo of the AI Futures Project warned the deal could end as 'safety paperwork' rather than a real slowdown, drawing an unflattering parallel to the social-media era's regulatory playbook.
Underneath is a concrete fear that sets the whole conversation's deadline: recursive self-improvement, where systems train and improve themselves outside human oversight. Anthropic's own timeline puts that milestone in early 2027. Multiple researchers at Anthropic and OpenAI have said publicly they believe advanced AI could cause human extinction within years. Whether any of this survives contact with the China argument, that a US pause simply hands a lead to Beijing, is the next test.
Shared on Bluesky by 2 AI experts
-
For a lot of AI researchers, the past weekend drew attention to an issue they've been sounding the alarm on for years. “There’s been this growing chorus of voices saying, ‘Please don’t build superintelligence soon. We ar…
View on Bluesky →
Originally reported by theverge.com
Read the original article →Original headline: Is Big Tech’s AI slowdown a safety pact or a cartel?