/Tech2h ago

Prime Intellect's Will Brown says third-party vendors bypass distillation bans by selling competitor-generated training data to major labs

Story Overview

Will Brown of Prime Intellect described an indirect channel in which US third-party vendors reportedly generate training tasks with Claude and Codex before selling the resulting data to both OpenAI and Anthropic, presenting the arrangement as a way around direct distillation restrictions.

338131.7K

#573

Original post

will brown@willccbb#573inTech

@xlr8harder in the US we have the more elegant mechanism where neutral 3rd party vendors use both Claude Code and Codex to help create training tasks to sell to both OpenAI and Anthropic which definitely doesn’t count as distillation

xlr8harder@xlr8harder

I completely believe Chinese labs are "distilling" US models at scale, and I have no problem with it whatsoever and think they should keep it up.

The model output licensing frame US labs want is bad and anti-human.

4:12 AM · Jun 19, 2026 · 1.5K Views

Open Question

No named vendors or scale details appear

Brown’s post supplies the sole source for the claim, with no listed companies, contract volumes, or independent reports confirming the practice exists at any measurable level.

Policy Risk

Terms-of-service enforcement stays unresolved

Whether selling model-generated tasks between frontier labs violates existing distillation policies or acceptable-use rules has not been addressed by the companies involved.

Sentiment

Sentiment building, check back later.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS359LIKES1

xlr8harder@xlr8harder

@willccbb Hahaha

will brown@willccbb

1h35910

Vivek@vivek_2332

@willccbb @xlr8harder 😂😂

2h631

umumu@umi33563

@willccbb @xlr8harder Are you sure about this? (Because I convinced myself (not thanks to any kind of indepth knowledge or something) that Ant is extremely cautious about using any synth data created without their complete oversight.)

1h321