Sam Altman renews his ‘democratic AI’ pitch as Anthropic holds a harder line

The OpenAI chief casts the AI race as democracies versus authoritarians. Dario Amodei’s Anthropic wants tighter chip export controls, not looser ones.


Sam Altman renews his ‘democratic AI’ pitch as Anthropic holds a harder line Image by: Shutterstock

Sam Altman has a preferred way of framing the contest over artificial intelligence, and it is a stark one. Either the United States and its allies build the technology and spread its benefits, the OpenAI chief has argued, or authoritarian governments use it to cement and expand their own power.

That binary, set out most fully in a 2024 Washington Post op-ed and carried into his appearances before Congress since, is the spine of what he calls “democratic AI”.

A recent Business Insider piece places the argument beside a quieter friction with Anthropic, whose chief executive Dario Amodei has been pushing a US-led AI coalition on notably different terms. The two men agree on the threat. What separates them is the remedy, and increasingly the register in which they make it.

Altman’s prescription leans on scale and openness. He wants more data centres and the power to run them, clearer rules on foreign investment, and a democratic system that reaches allied nations before an authoritarian one does.

On openness, OpenAI has something to point to. Its gpt-oss models, released in August 2025, were the company’s first open-weight systems in years, a pair of models that anyone can download and run rather than access through a paid interface.

For the democratic-AI case, that matters, because open weights are the version of the technology that actually travels to allied developers and governments. The gesture still sat awkwardly against OpenAI’s absence from an industry letter backing American open models.

Altman has paired that with a call for shared rules. He has floated an international body to set AI norms, comparing it loosely to the organisation that governs internet domain names, and has told lawmakers the state should fund safety testing rather than gate models behind approvals.

Amodei has taken the harder line on the hardware underneath all of it. In a January 2025 essay written after China’s DeepSeek unsettled markets with a cheaply trained model, he argued that “well-enforced export controls are the only thing that can prevent China from getting millions of chips”.

A more efficient model, he wrote, was no reason to loosen the rules. If anything, he read DeepSeek as evidence that the controls were doing their job.Sam Altman casts the AI race as democracies versus authoritarians. Dario Amodei’s Anthropic wants tighter chip export controls, not looser ones.

The backdrop is a US regime of chip export controls that has tightened repeatedly, with officials moving to close the loopholes that let advanced processors reach Chinese firms through overseas subsidiaries. Amodei sits at the hawkish end of that debate, framing strict controls as the difference between a unipolar and a bipolar world, language that treats the flow of chips as close to the whole game.

Altman’s framing is broader and, his critics would say, more convenient for OpenAI. His “commercial diplomacy” pitch pairs export controls with the active export of American AI to friendly states, which is also a fair description of OpenAI’s addressable market. He has said the values a technology carries travel with it, and that spreading a democratic version abroad is itself a security measure.

Anthropic’s bet runs the other way. Keep the most powerful hardware scarce, keep the frontier concentrated among a handful of accountable labs, and treat wide distribution as a risk to be managed rather than a race to be won. It is a safety argument and a competitive one at the same time, and Amodei has rarely pretended otherwise.

Neither posture is free of self-interest, and Business Insider is careful to note that the sharper contrast is partly the observer’s. Altman benefits from a world hungry for American models, and Anthropic benefits from one where being cautious and hard to copy counts as a moat.

What is clear is that the two most closely watched labs in the country now describe the same race in different vocabularies. One talks about spreading AI in order to win it. The other talks about withholding the chips that make it possible.

Get the TNW newsletter

Get the most important tech news in your inbox each week.