The leading figures of the artificial intelligence industry rarely agree on much, so when a consensus suddenly emerges among fierce rivals, it naturally captures widespread attention. Anthropic’s Dario Amodei, OpenAI’s Sam Altman, SpaceXAI’s Elon Musk, and other prominent leaders have stated in recent days that AI development needs to slow down. This alignment followed the resignation of an Anthropic researcher who issued a dire warning regarding the threats the technology poses to humanity.
Yet obstacles including domestic and global competition, profit motivations, and pushback from President Donald Trump stand in the way of a coordinated effort to pace AI’s development. Still, this moment could force the industry to wrestle with a reframing of what constitutes success, according to Nick Reese, former director of emerging technology policy at the Department of Homeland Security and an adjunct professor at New York University.
Reese suggested considering success in AI the same way society views commercial airline operations — on time and without crashes over extended periods. While society displays zero tolerance for commercial airline crashes, significant AI failures are routinely tolerated.
In theory, leading AI startups could slow the progress of their models at any time. In practice, technological leaps have partly sprung from competition to build increasingly powerful models, all with the blessing of a White House determined not to lose its edge over China. Anthropic and OpenAI are also racing toward potentially record-breaking public offerings on Wall Street, though OpenAI stated it is holding off on going public this year while continuing safety efforts.
Consequently, industry proposals depend upon mutual collaboration and government regulation. Amodei recently published a lengthy essay outlining a plan for slowing advanced AI development. One key proposal calls for frontier labs to commit to giving ongoing, employee-like access to independent outside evaluators. Evaluator teams would have offices, access badges, and company laptops to monitor safety practices continuously.
Anthropic is unilaterally committing to this step now, Amodei wrote. Altman called it a great idea on social media and noted OpenAI will follow suit. Altman welcomed a federal framework setting safety standards for advanced AI, but stated the company did not need to wait for legislation or an antitrust exemption before addressing safety.
Amodei’s essay also called for government regulation and intervention, including coordination between frontier AI companies with the help of the US government and other democratic nations. He expressed hope that leading labs could establish common safety standards as well as limits on unchecked AI progress. Furthermore, he suggested coordinating with authoritarian governments, acknowledging the difficulty of securing cooperation from China.
The Anthropic head outlined several levels of potential global agreement. The most feasible would ban narrow and obviously dangerous uses of AI, such as producing biological weapons or enabling users to do so. A more difficult level would involve the US and its adversaries testing models before release, possibly through a global standards body, for risks in cybersecurity or biological threats.
Still harder would be setting a speed limit on models capable of recursive self-improvement — developing improved versions of themselves on their own. Amodei noted that slowing the rate from extremely fast to only somewhat fast surrenders relatively little strategic advantage while potentially improving safety. He drew an analogy to Cold War treaties, where capping the number of missiles limited destruction potential while preserving each country's strategic deterrent.
Amodei identified the most difficult proposal as a full pacing or pause, where participating governments agree to substantially limit the overall rate of AI development. On social media, Altman clarified that pacing does not mean stopping. AI progress will continue to be rapid, Altman wrote on X, but should be slower than it otherwise could be. While safety cases and monitoring carry high costs, Altman stressed that pacing is well worth the cost, adding that no amount of American competitive pressure should justify recklessness.
While experts see the industry-wide agreement as promising, many unanswered questions remain. Sandra Wachter, professor of technology and regulation at the Oxford Internet Institute, noted that while a slowdown is possible in theory, in practice it requires substantial coordination between companies and governments across countries — something she deemed highly unrealistic due to the current state of the world.
Wachter suggested governments could promote a slowdown by holding AI companies responsible for risks, or by pursuing regulation that limits access to electricity, water, and other resources powering data centres.
Though embedding external evaluators inside companies has been mostly well-received, critics questioned assessor independence and evaluation standards. Aidan Gomez, co-founder and CEO of Canadian AI lab Cohere, warned that a handful of the most powerful labs based in one country agreeing on shared standards and advancement limits could further concentrate power.
Gomez also pushed back against Amodei's idea to have the US government give frontier labs waivers to bypass antitrust restrictions. If this is the most consequential technology in human history, Gomez stated, the rules cannot be written by a small group of commercially aligned companies behind an antitrust waiver.
Elham Tabassi, director of the AI and Emerging Technology Initiative and senior fellow at the Brookings Institution, said until commitments are solidified and public, independent auditors remain a voluntarily provided, company-controlled move. Tabassi emphasised that AI companies must continue investing in scientifically valid ways of testing and measuring models, warning that regulation will not hold power without them.
The Trump administration has long favoured a light-touch approach to the AI industry, hoping innovation flourishes and US development surpasses China. Still, some industry leaders remain cautiously hopeful about AI development moving in a safety-minded direction. However, while the issue is receiving attention, attention is not the same as implementation, leaving open whether it will really work.