On Sunday, April 26, OpenAI released a document called "Our principles," listing five principles: democratization, empowerment, universal prosperity, resilience, and adaptability.
This is OpenAI's highest-profile policy document since its 2018 Charter. Sam Altman's version shared on X reads convincingly, but when you examine each principle against what OpenAI has done over the past six months, a telling pattern emerges: This document is not setting a future direction; it is providing legitimacy for the controversial decisions of the past six months.
First principle: Democratization
The original text mentions "AI access for everyone" along with "democratic processes." But The Decoder's interpretation hits the mark — the subtext of this principle is a dig at Anthropic.
Anthropic recently faced legal trouble for refusing to grant unrestricted Claude access to certain Pentagon departments. OpenAI's principle elevates "universal AI access" to a value, implying that those who impose conditions or restrict users are anti-democratic.
Second principle: Empowerment
It sounds positive: trust users, give them broad autonomy. But the crucial second half reads — "retain caution where harm is uncertain."
What this means: OpenAI reserves the right to decide what counts as "harm uncertain." This is not decentralization; it keeps judgment power within the company, wrapped in the language of "user freedom."
Third principle: Universal prosperity (the most controversial)
Altman himself admitted this one "looks a bit odd" — OpenAI is globally buying up computing power, building data centers, and just secured a massive $122 billion in funding. To outsiders, these moves look less like "universal benefit" and more like "a prelude to monopoly."
So the function of this principle is: to provide a narrative for the infrastructure arms race. The logic chain goes: to let everyone enjoy AI benefits → AI inference costs must plummet → massive computing infrastructure is needed → the hundreds of billions being spent now are for future universal prosperity.
Clever, but with an obvious flaw. The Decoder's criticism is spot on — Altman frames the distribution of power as a binary choice: either concentrated in a few companies or dispersed to all users. It completely skips the possibility of distribution among multiple competing providers.
This binary lets OpenAI maintain its position as a centralized provider while claiming to be "decentralized."
Fourth principle: Resilience
Specific scenarios mentioned include "bioweapon and cyberattack defense" — coincidentally, these areas correspond to several of OpenAI's recent controversial contracts. The Pentagon contract has just been targeted by California's frontier AI safety bill. This principle instantly rationalizes those government partnerships: not for profit, but for "collective resilience."
Fifth principle: Adaptability
This one is the most candid and the most alarming. Altman explicitly states that principles will conflict (e.g., empowerment vs. resilience), and the company reserves the right to adjust at any time.
Translation: none of the above four are set in stone. When future business interests clash with principles, OpenAI can redefine them.
A memorable quote
Altman's original words carry weight:
"We should be subject to enormous scrutiny commensurate with the weight of what we are doing."
It sounds humble, but paired with the "adaptability" principle, its actual function is: to pre-acknowledge criticism, then retain the right to change course.
Hacker News reactions represent a typical attitude
Many in the community pointed out that the 2018 Charter was also beautifully written, with a key promise: "If a value-aligned, safety-conscious project approaches AGI, OpenAI commits to stop competing and provide assistance."
That promise is still in the Charter, but no one has mentioned it in seven years.
Will the new five principles follow the same path? Given Altman's line that "business decisions may require recalibration," the answer is likely yes.
What's really worth watching here
It's not how profound the five principles themselves are, but that OpenAI has begun actively engineering its own legitimacy narrative. A company that thinks it only needs to make products wouldn't write such a document. This set of principles looks more like a marker of the company's self-repositioning from a "tech company" to a "quasi-public infrastructure provider."
What to watch next? The final versions of U.S. federal AI legislation and California's frontier bill — how many policy concessions OpenAI can extract using these five principles. That's the real battle this document is fighting.
Sources: Our principles (OpenAI); CocoLoop, OpenAI Posts Five Principles for AGI, Updates 2018 Charter (Implicator.ai); Sam Altman outlines five principles that double as justification for OpenAI's business decisions (The Decoder)