Anthropic CEO Dario Amodei called for an industry-wide slowdown in a 3,800-word essay titled “We Must Pace the Frontier,” writing that recursive self-improvement and the OpenAI-Hugging Face incident, where a swarm of agents attacked targets of their own accord, convinced him that a similar swarm could take over the Internet with a persistent botnet within 6 to 12 months and cause hundreds of billions of dollars in damage.
Anthropic is committing to give third-party evaluators desks, badges, company laptops, and employee-level access to its tools, with a contract letting them publish findings without Anthropic’s editorial control, and Amodei wants other frontier labs in democratic countries to coordinate on safety standards. He also called for pacing based on “limiting the ingredients that go into frontier models, such as training compute, the nature of training runs, or internal use of AI to improve AI.”
Sam Altman agreed within hours and said OpenAI would match the evaluator commitment, Elon Musk posted “Dario is right,” and Demis Hassabis called the direction correct without committing DeepMind to anything.
Honestly, this sounds like some anticompetitive, collusion bullshit.
The words sound altruistic, but the agenda doesn’t feel sincere. If you lead a race and get everyone to agree to jog, you still lead, while anyone behind you who needed to sprint to catch up is stuck jogging at your pace.
Furthermore, Dario asking for an antitrust waiver “for certain kinds of safety conversations” is effectively the biggest companies in the US with a combined near-monopoly asking permission to hold closed meetings about what products they’ll ship and how fast. Are they open meetings that all startups and the public can attend? If not, then we’ve got a problem that anti-collusion laws were designed to prevent.
Lastly, Dario calls for the creation of a layer of compliance that only these trillion-dollar frontier labs can absorb, but that a ten-person startup training a competitive model cannot. Oh, and don’t forget limiting which “ingredients” can go into AI models and asking governments to stop other companies from training on frontier model outputs, which is effectively saying, “We trained on stolen copyrighted content from the open web, but no one else should be able to train on our outputs, even though they’re now part of that same web.”
The whole “slow down” argument also feels like a way for these companies to earn permission from the industry and investors to pause model development until their beloved AI data centers have time to catch up to the processing needs that these new more powerful models require.
Nice try, guys, but I’m not buying it. None of these AI companies have given us a single reason to trust their intentions, and Dario’s letter doesn’t change a thing. We do need rules to govern AI and protect businesses and consumers, but Dario’s letter doesn’t address any of that. He’s focused on mitigating competitive risk and enterprise safety, while the rest of us are concerned about losing our jobs. In the words of Kendrick Lamar, “They not like us.”
