BNN Bloomberg Sep 18, 2026
With Eddie Yoon, co-founder of Category Pirates
Asked what regulation the AI companies will face, Eddie Yoon answered with a recall from 1982.
The standard framing of the current safety debate is that the technology has outrun the law. Yoon's position is that the law on outcomes is already written, that regulating how a model is built is pointless, and that the calls for a slowdown are being made by companies with listings to protect.
"Nobody wants a busted IPO. That is the main thing on the table. That is really the question behind the question."
Yoon co-founded Category Pirates and came on the program to talk about what the AI companies delaying their listings are actually doing. Eddie Yoon of Category Pirates on BNN Bloomberg.
The full segment is covered here so you can skip it.
Here are the 5 arguments that matter.
Key Takeaways
The existing laws already protect consumers, on his reading, which is why he thinks new technical rules are unnecessary
He wants the frontier labs to behave the way Tylenol's chief executive did in 1982 — a recall taken against the FBI's and the FDA's advice
The motive behind the slowdown calls is the listing: he says a failed IPO is the thing nobody wants
Buyers want data, a decision and damage control — and he says the decision is the hard part because humans produce gridlock
Regulate outcomes, not inputs: product-liability law already applies to a model the same way it applies to a vacuum cleaner
1. The Tylenol Precedent
Asked what rules AI companies will face over the next few years, Yoon said it depends on how the companies handle themselves, and framed heavy regulation as a competitive risk rather than a safety measure.
His argument against writing new rules
You don't want regulations to go wild, because then what you will end up with is China winning the AI race ahead of the US.
Eddie Yoon
Because the statute book already covers it
most of the laws that are already in place do the job of protecting consumers
Eddie Yoon
The case study he reached for is the 1982 Tylenol poisonings, when seven people died from capsules laced with cyanide. Yoon described what the company's chief executive, James Burke, did about it.
The recall he holds up as the standard
he pulled all the bottles off the shelf, took a massive market share hit, one to two billion dollars in market cap, but it was the right thing to do, even though the FBI and the FDA said not to do it that way
Eddie Yoon
Burke did it on principle to keep consumers safe, Yoon said, and it is taught today as a case study in crisis leadership. The application to the present is direct, and he does not think the labs have earned the comparison.
What would make new rules unnecessary
And if the frontier labs demonstrate that same capability, which they haven't yet done so far, they will actually be fine, and the industry can regulate itself
Eddie Yoon
A government body will not be able to understand what is going on in any case, he added, because the frontier labs do not fully understand it either.
2. The IPO Behind the Ask
The host asked whether the calls for a slowdown from the largest players are about safety or about locking out smaller rivals. Yoon declined to choose.
He allows both motives
Well, more than one thing can be true.
Eddie Yoon
There is an element of principle in it, he said — concern about humanity and how to serve the planet better. Then he named the incentive.
The line he used to get to the real motive
And, show me the incentive and I'll show you the outcome.
Eddie Yoon
And the motive itself
Nobody wants a busted IPO. That is the main thing on the table. That is really the question behind the question.
Eddie Yoon
His reference point was SpaceX, which he said listed first and did so successfully. What is unresolved for the AI companies, on his account, is whether the financials are coming in lower, whether cash flow is thin or negative because of the rate at which they are spending, or whether the risk disclosure now required in a prospectus has expanded because of the claimed chance that the technology destroys humanity. He said it is not clear which, and that the common factor is that nobody wants a failed listing.
The schedule as he understands it
Sam has already delayed the OpenAI IPO until 2027.
Eddie Yoon
What Anthropic's Dario Amodei does next is still open, he said.
3. Data, Decisions, Damage
The host noted that Anthropic is calling for a slowdown while pushing ahead with what she put at a $2 trillion listing, and asked whether the position damages its valuation. Yoon said it does not have to, and turned it into a question about how the two companies present themselves.
He objects to the word "labs"
There's a difference between labs and R&D and engineering and product delivery in a real company.
Eddie Yoon
He contrasted the frontier labs with Satya Nadella at Microsoft and Marc Benioff at Salesforce, whom he described as running companies where a product is tested and evaluated and the risk is accounted for before it ships. What a corporate buyer is actually buying, he said, breaks into three parts.
The first is the data, which he called the standout value of a large language model. The second is the decision, and that is the one he says the industry underrates.
Why the decision does not follow from the output
Decision-making is actually quite hard, and it's largely controlled by humans, not one human, but many humans, and the larger the organization, the more humans are involved, and humans tend to lead to gridlock.
Eddie Yoon
The third is damage control, and he put it in purchasing terms.
What the buyer is looking for
what companies actually want is one throat to choke
Eddie Yoon
If something goes wrong, he said, customers want to know that Altman and Amodei will do what Burke did in 1982, even at the cost of their own profits.
4. Lying With the Truth
On OpenAI's stated reason for delaying — that it needs the freedom to make safety choices that may hurt short-term profits — the host asked whether Yoon buys it.
His answer was both at once
Sam is telling the truth, and he might actually be lying a bit with the truth.
Eddie Yoon
Safety genuinely matters and OpenAI wants it handled, he said. It is also true that outsiders cannot see the monthly revenue run rate, and that a slowing rate is bad for a listing, as is failing to show positive gross margins.
How much of the hand is visible
you're only seeing the first cards of a Texas Hold'em play out, and you have not seen the flop or the river yet
Eddie Yoon
5. Outcomes, Not Inputs
The host's last question was whether competition with China makes a voluntary slowdown impossible, and whether rules only work if every country signs up. Yoon split regulation in two.
One kind governs the technical inputs — the equivalent of banning an ingredient in a hot dog, or in a model. The other governs outcomes, and specifically the unintended ones. On outcomes, he said, the rules already exist: if a model caused mass hysteria leading to death and destruction, there is ample law covering it.
He cited Lina Khan, the former chair of the Federal Trade Commission under the Biden administration, as having made the same point about product-liability law: a product that causes death or destruction leaves its maker accountable. Yoon said that holds regardless of what the product is.
The scope of product liability
that's true, whether it's an LLM or a Roomba or a Keurig, it doesn't really matter what it is
Eddie Yoon
That is also, on his argument, the part of the problem that does not need international agreement, because nobody running an open-source model in China wants to be blamed for deaths caused by it either.
His verdict on the other kind of rule
Regulating the inputs and the technical aspects is a waste of time.
Eddie Yoon
Bonus Insights
The host opened by saying major AI companies are delaying their listings amid calls from technology leaders to slow development down, which is the framing Yoon spent the segment disagreeing with.
The $2 trillion figure for Anthropic came from the show, not the guest. The host put it into her question and noted there is no word on timing.
Yoon was introduced under the wrong name and firm, and the host apologized on air once the graphic showed his actual name, before restarting the interview.
The host conceded the case for more rules while framing the obstacle as coordination: she said few people would argue against more AI regulation given the pace of development, but that it needs every country on board to work.
Yoon's bottom line is that the current safety argument is a listing argument in disguise, and that the useful regulatory work is already done: hold a company liable for what its product does, and leave the question of how the model was built alone.
Products, Companies & Tools Mentioned
OpenAI (Has delayed its listing to 2027, on Yoon's account, with the run rate and gross margins the unknowns behind the decision)
Anthropic (Calling for a slowdown while pursuing a listing the host put at $2 trillion; Yoon says the position need not cost it valuation)
Tylenol (The 1982 cyanide poisonings and the recall that followed, which Yoon uses as his standard for how a company should behave in a crisis)
Microsoft and Salesforce (Named as the counter-example: chief executives who ship a product only after it has been tested and the risk accounted for)
SpaceX (His example of the listing that got out first and went well, against which the AI companies are now being measured)
iRobot's Roomba (His illustration that product-liability law applies to a model exactly as it applies to a household appliance)
Category Pirates (The newsletter he co-founded, and the credential he appeared under)
Watch the full episode:
More Eddie Yoon interviews and podcast appearances
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: