Bloomberg Tech

Neil Chilson on why a safety waiver risks a durable AI cartel

Neil Chilson· Head of AI Policy at Abundance Institute at Abundance Institute
·~6 min·English·Bloomberg
PolicyAI SafetyBusiness Strategy
TL;DR

A former FTC chief technologist argues frontier labs do not need an antitrust waiver to build AI safely, and warns the exemption they want is exactly what turns a fragile cartel into a durable, government-backed one.

01The Ask

Labs already have room to cooperate

<strong>Frontier labs already have many legal ways to coordinate on safety</strong>, so the antitrust waiver they want is about legal cover, not a missing capability.

So there's so many ways that companies can work together that are not anticompetitive.

— Neil Chilson, Bloomberg Tech
Key Insight
His answer implies a sharp test for the waiver: name one safety task the labs could not already do together lawfully. If none exists, the exemption is buying legal insulation for the coordination, not the ability to coordinate.
02Core Mental Model

Only government can make a cartel durable

<strong>Cartels normally collapse because members defect to compete</strong>, so the only durable ones are those a government protects, which is the risk Chilson sees in the exemption the labs want.

The only cartels that are durable are ones that are endorsed by the government.

— Neil Chilson, Bloomberg Tech
Key Insight
The mechanism he leaves implicit: private cartels crack because any member can defect and there is no lawful way to stop them. An exemption does not remove the temptation to defect, it can legalize the coordination and the means to police it, which is why scope and duration are the whole game, broad and lasting cover could hold a group together long after its safety rationale fades.
03The Safety Net

The law already covers the harm

<strong>If a model harms a third party, existing FTC consumer-protection and tort law already reach the company</strong>, so a legal safety net is in place before any AI-specific law is written.

the law is pretty clear, especially FTC consumer protection law, and I think many tort laws, that those companies will be responsible for those harms.

— Neil Chilson, Bloomberg Tech
Key Insight
The argument leans on a quiet condition: liability only deters if responsibility can actually be pinned on a lab. Where a model's harm is hard to trace back to its maker, the law can be on the books and still change very little about how carefully the model is built.
04Inside the Regulator

The FTC would judge this on two fronts at once

<strong>As a former FTC chief technologist, Chilson describes the agency weighing AI on two distinct axes at once</strong>: whether the practices serve competition and consumer welfare, and, separately, whether they are unfair or deceptive to consumers.

they would be looking at the various authorities that they have both on the competition side and as well on the consumer protection side.

— Neil Chilson, Bloomberg Tech
Key Insight
The reasonableness test makes a lab's own safety record legally load-bearing even without new AI law. Could this harm have been easily avoided turns internal safety practices into potential evidence, not just good hygiene.
05Early Warning

Small incidents are the alarm, not the fire

<strong>Chilson hopes catastrophic risks will surface first as small, low-harm incidents</strong>, and reads today's red flags as a warning system that buys time to correct course.

there's time to look at the laws, to look at the practices in the industry, and say, like, hey. We need to change course here.

— Neil Chilson, Bloomberg Tech
Key Insight
Reading minor incidents as an early-warning signal rather than a failure is what lets him favor iterative correction, filling legal gaps as they appear, over pre-emptive sweeping rules. But it rests on two conditions he cannot guarantee: that the worst harms announce themselves in small ways first, and that institutions act before the window closes.
06The Balance

Mitigate the harm without killing the upside

<strong>The same unfairness test that catches harm also weighs it against benefits</strong>, and Chilson insists the enormous upside of AI must not be lost to over-cautious mitigation.

there are enormous benefits. We're seeing that in stock values. We're seeing that investment, and we're seeing it in the huge amount of adoption and usage that individuals are doing.

— Neil Chilson, Bloomberg Tech
Key Insight
There is a gap a careful reader will catch. The unfairness test weighs a specific practice's harms against that same practice's benefits, but the benefits Chilson reaches for, stock values, investment, adoption, are aggregate. Enormous benefits from AI in general do not automatically clear any one harmful practice.