Politics

Copy article

Altman Announces OpenAIs 5 Principles. How Do They Compare to the 2018 Charter?

ended 27. April 2026

On April 26, Sam Altman, OpenAI CEO, just rewrote its constitution. Not a tweak. A replacement. The 2018 Charter, the document that defined what OpenAI owed the public, has been superseded by a new set of “principles”. The company hasn't withdrawn the old Charter. It's still sitting at openai.com/charter. But the new document reads like it was written for a different organisation. In important ways, it was. Here's what OpenAI promised then, and what it says now. 

Some of what's new and positive is genuinely worth noting. The 2026 principles are broader in scope, more honest about uncertainty, and more explicit about the role of governments and society in shaping AI's future. The commitment to resist power consolidation is stated plainly. The acknowledgment that AI introduces real risks,  including biosecurity and cybersecurity threats, is concrete and specific. And the sentence "we are committed to doing our part to make the future better than the past" is, at minimum, a stated obligation.

But the new document stops saying things, and what OpenAI stopped saying matters as much as what it started. Here's what OpenAI promised then, and what it says now. Side by side.

On duty:

2018: "Our primary fiduciary duty is to humanity."

2026: [Deleted. No replacement.]

What this means: The only sentence in OpenAI's founding documents that named a specific obligation to anyone outside the company is gone. No equivalent constraint exists in the new principles. The public has gone from named beneficiary to unnamed bystander.

On competition:

2018: "We are concerned about late-stage AGI development becoming a competitive race without time for adequate safety precautions. Therefore, if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project."

2026: "We can imagine periods in the future where we have to trade off some empowerment for more resilience."

What this means: The stop-and-assist clause was the only structural brake on an AGI arms race written into any major lab's charter. In its place: a willingness to collaborate when OpenAI judges the moment requires it. That's not nothing. But a self-assessed pause is a different instrument from a triggered obligation. The decision about when to slow down moved from a defined condition to a judgment call made internally.

On conflicts of interest:

2018: "We will always diligently act to minimise  conflicts of interest among our employees and stakeholders that could compromise broad benefit."

2026: [Deleted. No replacement.]

What this means: This deletion arrived after OpenAI converted to a for-profit PBC, after Microsoft took a $135 billion stake, and after investor profit caps were removed. The sentence that acknowledged the tension between private gain and public mission disappeared at the exact point that tension became structural.

On power concentration:

2018: "We commit to use any influence we obtain over AGI's deployment to ensure it is used for the benefit of all, and to avoid enabling uses of AI or AGI that harm humanity or unduly concentrate power."

2026: "We will resist the potential of this technology to consolidate power in the hands of the few."

What this means: This is one of the few places where the 2026 language is comparably direct. "We will resist" is a commitment verb. The question is whether "resist the potential" carries the same weight as "avoid enabling." One describes an intention. The other described a constraint on behaviour. But the principle survived, which is more than can be said for several others.

On commitment language overall:

2018: "We commit." "We will." "We expect." used throughout as default register.

2026: "We believe." "We envision." "We can imagine."  now the dominant register, with selective use of "we will" for specific pledges (transparency, resilience, resisting power consolidation).

What this means: The 2026 document does contain committal verbs, but they've shifted from the centre of gravity to the edges. The overall register moved from obligation to aspiration. You can hold someone to "we commit." It's harder to hold anyone to "we believe." Where the 2026 document does commit, it commits to process (transparency, collaboration). Where the 2018 document committed, it committed to constraints on the company's own power.

On who acts:

2018: "We commit to use any influence we obtain..."

2026: "Our governments may need to consider new economic models..."

What this means: The subject changed. In 2018, OpenAI was the actor and the constrained party. In 2026, global governments and society share the stage, which is arguably more realistic, but also distributes responsibility at the exact moment OpenAI shed its own. Asking governments to build the safety net is reasonable. Doing it in the same document that deletes your own obligations is worth noticing.

What's genuinely new and worth acknowledging:

The 2026 document names specific risks, bioweapons, cybersecurity, that the 2018 Charter didn't. It commits Foundation resources to addressing them. It acknowledges that OpenAI is now too large to pretend it's a scrappy research lab. And the principle of adaptability, being willing to say “we got this wrong”, is more honest than most corporate documents manage, especially amongst the AI booster community.

OpenAI completed its conversion to a for-profit public benefit corporation. Microsoft acquired a £100bn ($135bn) stake. The nonprofit Foundation received $130 billion, just below Microsoft's holding. OpenAI removed all safety language from its IRS filing released in November 2025. Fortune reported six mission statement changes in nine years. 

Elon Musk's lawsuit, which specifically challenges whether OpenAI breached its fiduciary duty to the public, is heading to trial after six years. Jury selection starts today, Monday 27 April, in the US District Court for the Northern District of California. Opening arguments are expected to begin Tuesday, with Yahoo News reporting Musk potentially taking the stand as early as that day.

The words changed. The question is whether the obligations did, or whether the obligations changed first, and the words just caught up.

We'd like your views:

  • When a company replaces binding commitments with aspirational beliefs at the same moment it restructures for private investment, should regulators treat the new document as governance or as marketing?
  • The 2018 Charter committed OpenAI to stop competing and assist any rival closer to safe AGI. That clause is gone. Does removing a cooperation pledge during an arms race make the race safer or more dangerous?
  • OpenAI now says governments should consider new economic models to share AI's benefits. Is that democratisation, or is it outsourcing the obligation the company just deleted from its own charter?
  • If "our primary fiduciary duty is to humanity" was important enough to write down in 2018, what changed about humanity's interests that made it deletable in 2026?

4 responses from the Newspage community

Copy all

Star Quote
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
OpenAI says governments need to step up, but no government in the middle of an arms race wants to be the one that slows down. The competitive pressure between labs is mirrored between nations. Everyone is running, nobody is steering. Calling for government action while the industry makes that action politically impossible is a neat trick.

Even for pro-AI nations, it's brutal. In the UK, around 140 data centres are seeking grid connections needing 50 gigawatts. Britain's entire winter peak spike in demand is 45 gigawatts.

Communities are fighting these builds through planning courts. The first judicial review of a hyperscale data centre's environmental impact landed in January. Becoming a country that supports AI infrastructure isn't a simple policy announcement. It entails a decade of planning battles, grid upgrades, and water supply negotiations with local people who have every right to ask what's in it for them.
Copy

Editor
Katrina Young
Futurist | Chief Technology Officer CTO/CDO Ethical AI & Digital Transformation Strategist at KYC Digital
OpenAI’s updated principles arrive as a US court examines its original commitments. The timing is notable, the real shift is structural. The 2018 Charter was a governance document. It defined obligations and constrained behaviour. The new principles operate differently. The language moves from commitment to intent. That is not stylistic. Commitments can be tested. Principles require interpretation. What matters is not just what was added, but what changed in form. The Charter included a cooperation clause acting as a brake in an AGI race. That mechanism no longer appears in the same way. Decisions on pacing and collaboration now sit within organisational judgment.

This reflects a broader reality. OpenAI is no longer a research lab in isolation. It is a scaled, commercial actor operating across governments and markets.
The risk is not principles themselves. It is the absence of independent standards to assess how they are applied. When accountability relies on internal interpretation.
Copy

Rohit Parmar-Mistry
Founder at Pattrn Data
When a company drops binding commitments and replaces them with broad principles at the same time it restructures around private capital, nobody should confuse that with neutral governance. That is a power shift. The language matters because charters are where lofty intent meets operational restraint, and what has disappeared from OpenAI's framing is just as important as what remains.

The old promise to step back and cooperate if another lab got closer to safe AGI was extraordinary. Removing it during an arms race does not make the system more stable. It signals that competitive pressure has beaten institutional humility. In plain terms, the mission has moved closer to market logic.

If firms now want governments to spread the gains, while deleting stronger duties from their own constitutional documents, regulators should ask a simple question: which obligations are being socialised, and which rewards are being privatised?
Copy

Taryn Lee Johnston
Owner at The FCM Group
The shift from “we commit” to “we believe” is doing a lot of work here.

If binding commitments are replaced with aspirational language at the same moment a company restructures for private investment, it’s hard to treat that as governance. It reads much closer to positioning.

Removing the stop-and-assist clause is more significant than it looks. That was one of the few explicit brakes on competitive escalation. Replacing it with internal judgement doesn’t remove the risk, it concentrates it.

On the point about governments, there’s a valid role there. But shifting responsibility outward while reducing your own stated obligations creates an imbalance.

The underlying question isn’t whether the wording changed. It’s whether accountability moved with it.