I stumbled onto something fascinating this week. You know how everyone talks about OpenAI’s transformation from nonprofit to whatever-it-is-now? Turns out there’s a paper trail that tells the whole story, and it’s hiding in plain sight on their IRS tax returns.
Simon Willison did the legwork here (props to him for the detective work). He extracted the mission statements from OpenAI’s 501(c)(3) tax filings from 2016 through 2024 and turned them into a git repository. The result is basically a diff history of how a company’s values change when reality hits idealism in the face.
The Original Vision
The 2016 mission statement is almost quaint in retrospect. They wanted to advance “digital intelligence” (remember when we called it that?) in a way that would “benefit humanity as a whole, unconstrained by a need to generate financial return.”
But here’s the part that really gets me. They explicitly said they were “trying to build AI as part of a larger community” and wanted to “openly share our plans and capabilities along the way.”
That lasted exactly two years.
The Slow Retreat from Openness
By 2018, the community collaboration bit was gone. Just deleted. No explanation, just poof. This was before GPT-3, before ChatGPT, before any of the massive breakthroughs that supposedly justified the pivot to secrecy. They were already backing away from the “open” in OpenAI before they had anything valuable enough to protect.
The 2020 changes were subtle but telling. “Benefit humanity as a whole” became just “benefit humanity.” Dropping “as a whole” might seem minor, but it changes the scope. You can benefit humanity without benefiting all of it equally. It’s a hedge, linguistically speaking.
They kept repeating that “unconstrained by a need to generate financial return” line like a mantra though. That stayed in there through 2022, even as they were already structuring deals with Microsoft and pivoting toward the for-profit subsidiary model.
When Confidence Replaces Humility
The 2021 revision is where things get really interesting. “Digital intelligence” became “general-purpose artificial intelligence,” which is fair enough given how the terminology evolved. But they also changed “most likely to benefit humanity” to just “benefits humanity.”
That’s not a small shift. They went from acknowledging uncertainty to stating outcomes as fait accompli. It’s the difference between “we’re trying to do good” and “what we’re doing is good.” One invites scrutiny, the other assumes righteousness.
They also stopped wanting to “help the world build safe AI technology” and started focusing on developing it themselves. The collaborative framing disappeared entirely. Now it was about what “the company” would do, not what they’d enable others to do.
The 2024 Nuclear Option
And then 2024 happened. They nuked almost everything and reduced it to one sentence: “OpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity.”
Gone is any mention of safety. Gone is the “unconstrained by financial return” language. Gone is basically everything except the vaguest possible commitment to doing something good for everyone at some point maybe.
It’s actually impressive how much you can say by saying less. This new version doesn’t commit them to anything specific. It doesn’t explain how they’ll ensure AGI benefits everyone. It doesn’t acknowledge tradeoffs or competing interests. It’s just corporate poetry that sounds nice in a press release but means nothing in practice.
What Tax Filings Tell Us
Here’s why this matters beyond just dunking on OpenAI for being hypocritical. These aren’t blog posts or tweets that some PR person drafted. These are legal documents filed with the IRS. The mission statement on a 501(c)(3) tax return has actual regulatory weight. The IRS can use it to evaluate whether an organization deserves to maintain its tax-exempt status.
Every change here represents a deliberate legal and strategic decision. Someone sat in a room and decided which words to keep and which to delete. They knew these changes would become public record. They did it anyway because the alternative was keeping promises they couldn’t or wouldn’t keep.
I keep thinking about that 2018 deletion of “openly share our plans and capabilities.” That was the canary in the coal mine. Everything that followed was just elaborating on that initial betrayal of their founding principles.
The irony is that OpenAI might actually be doing important work on AGI safety and alignment. But by systematically removing accountability from their mission statements, they’ve made it harder to evaluate whether they’re succeeding or just saying they are. When your stated mission becomes indistinguishable from meaningless corporate speak, you’ve lost the ability to hold yourself accountable even if you wanted to.