If frontier labs coordinate a slowdown, when does “AI safety” become a cartel?

I understand the safety case for slowing down But if OpenAI, Anthropic and xAI actually agree with each other not to release more capable models, not to scale compute past some threshold, or not to move until the others are ready, that starts sounding less like safety policy and more like competitors coordinating to restrict innovation Imagine Apple, Samsung and Google agreeing that phones are advancing too fast so nobody will release a better one next year People would immediately ask antitrust questions Why should frontier AI be fundamentally different? If the slowdown is imposed by law, that is one thing If every lab independently decides to slow down, also fine But if Sam, Dario and Elon privately agree on capability ceilings or release delays, consumers get worse models later and competition gets weaker And the same companies that benefit economically from less competition get to decide how much progress everyone else is allowed to access Maybe such an agreement would genuinely reduce existential risk Maybe it would even be the right thing to do But “we are doing this for safety” cannot automatically mean “antitrust law no longer applies” At what point does coordinated AI safety become coordinated output restriction? Would love to hear from anyone who actually knows US antitrust law because this could become a very real issue surprisingly soon