The story of how Elon Musk left OpenAI has been told from many angles. Greg Brockman just told it from the inside.

His account, drawn from personal journals and testimony, reframes the 2017 split as something more fundamental than a disagreement about money or direction. It was about one question: who controls the most consequential AI lab in the world?

The Meeting

In late August 2017, OpenAI was transitioning to a for-profit subsidiary structure. This was the moment where the governance question became unavoidable — and where the founders and Musk finally had it out.

Musk demanded "unequivocal" control. The other founders proposed equal shares instead. Brockman says Musk "sat for several minutes thinking quietly" and then said: "I decline."

He stood up, stormed around the table, grabbed a painting, and started to leave. Then he asked: "When will you be departing OpenAI?"

It was a bluff — or close to one. But the message was clear: without control, Musk was done.

The Pattern Brockman Describes

Brockman's account adds texture that wasn't public before. He describes Musk's practice of giving Tesla Model 3s to co-founders as "buttering them up" ahead of governance negotiations. He says Musk dismissed an early demonstration of what became ChatGPT. And his verdict on Musk's relationship with AI is blunt: Musk "did not and does not know AI."

The word "certain failure" that Musk cited when leaving the board takes on new meaning in this context. Musk wasn't saying OpenAI would fail — he was saying OpenAI would succeed on a path he couldn't control. Those are different things.

The Timeline

  • August 2017: The critical governance meeting. Musk demands control, is refused, declines to participate.
  • Within 6 months: Musk stops regular donations to OpenAI's operating budget.
  • February 2018: Musk leaves the board voluntarily, citing Tesla as his AI focus.
  • 2020: Musk stops paying for shared office space with Neuralink.

The exit was gradual, not sudden — a slow withdrawal rather than a dramatic rupture. That pattern is consistent with someone who lost a negotiation and chose to reduce exposure rather than fight.

The Bigger Question

What makes Brockman's account significant isn't the drama — it's what it reveals about the governance of AI labs. OpenAI's unusual structure (nonprofit parent, capped-profit subsidiary) was designed specifically to prevent any one actor from having total control. Musk's demand for "unequivocal" authority was rejected not once but structurally, through the organization's design.

That design has held. OpenAI survived the split, the GPT releases, the governance crises of 2023-2024, and continues to operate. The structure worked exactly as intended.

Brockman's testimony is also a reminder that the people building AI systems make decisions about governance — and those decisions have long-term consequences. The for-profit conversion that triggered the Musk split eventually led to OpenAI's current structure, which in turn shaped how the most capable AI systems in the world are developed and deployed.

The AI race isn't just about compute and data. It's about governance. And OpenAI's governance is what allowed it to survive this moment.

Sources: TechCrunch