OpenAI and Anthropic Ask Washington to Be Able to Ban Dangerous AI Models

OpenAI and Anthropic have sent an open letter to the White House Office of Science and Technology Policy. It is an unusual move: two of the biggest AI companies are asking the state to control and regulate them more tightly.

What the letter says and asks for

The companies want government restrictions on the development and deployment of AI models that carry certain risks. In effect, OpenAI and Anthropic are asking that the US government have both the right and a mechanism to bar developers from deploying new models if they threaten security or national interests.

This is not a request for light-touch regulation or an industry self-code. The companies are asking for outright bans, where the state can say "no, you may not release this model." It resembles how nuclear technology, aviation, medicines or food additives are controlled: before mass production starts, a review and approval are required. Such an approach ensures that powerful systems do not fall into the hands of those who might use them to cause harm.

Why they are proposing this

Dario Amodei, Anthropic's chief executive, stressed that insufficient government oversight and weak vetting of customers could lead to serious threats to society. The company sees the problem in the fact that today anyone can get access to a powerful model without sufficient checks on how it will be used, for what purposes and who exactly will operate the system.

This is not only about marginal cases or theoretical scenarios. OpenAI and Anthropic look at an ecosystem of startups and companies that are growing fast and competing with each other in a race for funding and market share. Without government restrictions, that competition may lead someone to release a dangerous model without checks, simply to reach the market faster and attract investment. Haste is the enemy of safety.

The main risks that concern both companies:

• Use of models to create disinformation, manipulation and fakes at societal scale
• Deployment without verifying the customer's goals, integrity and actual intentions
• No mechanism for fines, bans and accountability for violations and leaks
• Leakage of technology to criminal organizations and hostile states
• Use for automated cyberattacks, phishing and breaches of critical infrastructure systems

The competition paradox and barriers to entry

An interesting economic paradox is at work here. Companies that have invested billions in R&D and reached global leadership in AI are asking the state to raise high barriers to entry for competitors. If government approval is required for every new model, small startups simply will not be able to compete: they lack the resources and the connections in Washington to get through a bureaucratic process. OpenAI and Anthropic, meanwhile, already have relationships with the authorities and may negotiate softer terms or faster review.

But this is not only a blatant business strategy. Amodei, Altman and the leadership of both companies appear to genuinely believe that very powerful AI requires government oversight. If they are right in that assessment, regulation would be logical, as it is for aviation, pharmaceuticals or nuclear energy.

What happens next

The White House's reaction is not yet known. But the letter arrives at a moment when the US administration is paying closer attention to AI policy and its potential risks.

If the authorities heed the appeal, the AI industry will change radically. Small startups will wait months or years for government approval. Releasing a new model will take longer. Rapid innovation may slow down. At the same time, the risk of uncontrolled release of truly dangerous systems, which could be used for disinformation or to cause harm, will fall.

The alternative is to continue the current wild-west competition and hope that companies' internal responsibility will be enough. History shows that in fast-growing, promising fields (social media, fintech, self-driving cars) such hope often proves misplaced. The market does not always allocate itself optimally for society.