The people building AGI want permission to collude, and they're dressing it up as safety.
The Summary
- Neil Chilson, former FTC Chief Technologist, warns that AI safety coordination could become a government-blessed cartel if labs get antitrust exemptions
- Companies can set safety standards without suspending competition law, he argues
- Existing consumer protection and liability frameworks already cover AI harms
The Signal
The frontier labs have a new pitch: let us coordinate on AI safety, and look the other way on antitrust. Chilson, who spent years as the FTC's top tech advisor, isn't buying it. His argument cuts through the safety theater: companies routinely collaborate on technical standards without needing antitrust exemptions, and AI should be no different.
This matters because the labs know exactly what they're doing. OpenAI, Anthropic, Google DeepMind, and others have spent two years building the narrative that AI is so dangerous and so special that normal rules shouldn't apply. First it was "we need to self-regulate before government steps in." Now it's "we need government protection to coordinate properly."
"The labs can set safety standards without suspending antitrust rules."
Here's what a government-backed AI cartel would actually look like:
- Incumbents coordinate on model release schedules, effectively controlling when new capabilities hit the market
- Shared "safety" standards become barriers to entry that smaller labs can't meet
- Computing requirements and training approaches get standardized in ways that favor those who already have the infrastructure
Chilson's point about existing law is the sharpest part of the argument. Consumer protection statutes, product liability, and fraud laws already cover the harms AI can cause. If a model hallucinates medical advice that hurts someone, that's a liability question. If it enables discrimination at scale, consumer protection law applies. The legal system isn't perfect, but it's also not blind to new technology.
The Implication
Watch what the labs actually ask for in the next six months. If they push for formal antitrust exemptions or industry-controlled certification regimes, you're watching cartel formation in real time. The right model is what already exists: companies can publish safety research, share best practices, and build common testing frameworks without coordinating on pricing, product releases, or market entry barriers.
For builders outside the frontier club, this debate determines whether you'll have room to compete. An AI safety cartel locks in today's winners. Open standards and normal antitrust enforcement keep the door open.