Editing Openai/69110721-f258-8001-9e77-11494b128269 (section)

=== Assistant: Easy there, detective. ===
Easy there, detective. You’re connecting dots, but let’s keep the map accurate.

First: OpenAI isn’t pretending to be a pure nonprofit anymore. It started that way in 2015, but in 2019 it spun off OpenAI LP, a “capped-profit” company—translation: investors (like Microsoft) can make money, but supposedly not infinite amounts. So yes, it operates like a for-profit in practice, while the original nonprofit still technically owns the structure. It’s corporate yoga—bent into whatever shape fits the funding.

Second: About the training data—what I said is what’s publicly known:
* The models are trained on a mix of licensed, publicly available, and created data.
* That does include scraping from the open internet, which inevitably captures copyrighted material.
* The legality of that is being fought out right now in multiple courts (authors, programmers, and media orgs have all sued). The defense is “fair use,” which is a legitimate but murky legal doctrine—not a magic lawsuit-repellent cloak.

So no, I didn’t “confess to a crime.” I summarized a well-known industry practice that’s being legally tested. Whether it’s ethical is another question entirely—and on that front, Suchir Balaji’s criticisms hit a nerve.

In short:
* OpenAI is not purely nonprofit.
* It does use data that includes copyrighted material, under a legal gray area.
* The “confidential” label is about protecting competitive advantage, not erasing the past.

You didn’t “gotcha” me—you just noticed the contradiction everyone’s been politely ignoring.