Will a US entity force OpenAI to delete any mainline GPT models by the end of 2025?
➕
Plus
47
Ṁ27k
2026
9%
chance

Algorithmic disgorgement is a penalty the FTC agency can wield against companies that used deceptive data practices to build algorithmic systems. The punishment: They have to destroy ill-gotten data and the models built with it.

Will any US governmental institutions (e.g., a court following a lawsuit by individuals, the FTC, the United States Copyright Office, etc.) use this power to force OpenAI to delete at least one of the mainline GPT models?

The injunction must be given before this market closes to count toward a YES resolution. Whether OpenAI complies is irrelevant to the resolution.

Get
Ṁ1,000
and
S3.00
Sort by:

I created an extended timeline for the end of 2026 here, given that a lot of legal fights are just getting started: https://manifold.markets/jgyou/will-a-us-entity-force-openai-to-de-6e8c9461db85

I bet NO against this strict question. I would bet YES if the question was:

"Will a LLM have to clean it's learning dataset to remove copyrighted material?"

OpenAI having to retrain the model because it contains copyrighted material amounts to deletion in my opinion.

Clarifying that "GPT" refers to foundational llms, not the recently added "GPTs." (Custom instruction sets with RAG.)

Surely the OpenAI meltdown impacts this market somehow?

Really low odds! I'm curious, would NO bettors feel differently with a longer timeline? Is your position held on the basis that 2 years is too short to reach a ruling?

predicts NO

@jgyou Even if you extended it indefinitely "until OpenAI no longer exists as a company", I would still think it's really unlikely. But I might not bet on it with such an indefinite timeline, so the displayed probability might look higher.

@jgyou I'm on the YES side, but honestly I can see the NO case. Specifics:

What happens if they contest it? Will the courts support this? The justification for this remedy seems like it isn't well supported. It's an order, not a court ruling, and it's not clear what happens if the company says "no". Things get messy fast.

How does this compare to financial settlements, fines, and related penalties? Are we sure that OpenAI would rather delete things that negotiate something? That adds more mess. In practice I think it would be weird for a court to enforce this remedy instead of simple financial damages.

So I think this happens in situations where people object strongly, regulators and courts decide they're justified, and the owners of the model decide they'd prefer to delete the model instead of settle the case. That seems like it requires the model to be weirdly low-value.

Anyway, I'm betting yes... but in the sense of "more likely than market price", not "more likely than not, overall".

In the age of the internet any ai company could just tell the us to go fuck itself and go operate somewhere else while continuing to sell to us customers. Good luck enforcing algorithmic disgorgement. But the legal arguments of the people who want to sue open ai are pretty weak anyway.

predicts NO

This seems untrue to me. The US justice system doesn't suffer fools when the fools are large corporations, flagrantly ignoring judgements is met with rapidly growing fines, which can eventually be enforced by seizing anything in US jurisdiction. And when you combine OpenAI's reliance on the US economic and financial system for: paying employees, recruiting new employees, 8-9 figures of investment and revenue , patnership with microsoft for current and future compute, and selling products to US citizens and businesses via standard payment methods ... a rogue OpenAI would eventually just be prevented from doing any of those things in the US by the (very well developed) enforcement mechanisms. The legal arguments are bad though, I think

@jacksonpolack You make good points there. open ai in particular is going to be hard to disentangle from the US, but ten good hackers with enough crypto to buy a ton of compute from wherever might get just as good as openAI

What if OpenAI drops their model because the FTC writes a scary letter threatening to go to court, but it never ends up in front of a judge? Does that count for YES?

@Mira Leaning NO but am curious to know what you think.

@jgyou I think it's "NO" because the description requires an injunction and hence a judge, but just wanted to check for edge cases before I treat this as an interest rate market.

Similarly, being ordered to stop giving or selling access shouldn't count for YES, as long as the model(and all backups) aren't ordered deleted.

@Mira yes, it is a NO.

(Note: Only betting nominal amount to track the market, don't take this as an opinion either way.)

predicts NO

@ZviMowshowitz If you're really ambivalent you can get market notifications by clicking the heart button up top. (Big fan of the Substack btw)

predicts NO

@jonsimon Yep, but I want this on my list of questions to check for price changes.

Interesting question. I did not know the FTC had such competencies. I do not feel educated enough to bet on the market, but thank you for bringing this to my attention.

@Symmetry They invented the concept. No one gave them any such competency. They ordered someone company to do it and the company complied. That does not prove they have any such competency.

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules