Sam Altman Wants a 'Superintelligence New Deal' — and OpenAI Has a Policy Blueprint to Match

Sam Altman went to Washington — metaphorically and literally — with a proposal that reframes AI safety as an economic justice issue. Published April 6 in Axios, his blueprint calls for a “Superintelligence New Deal” modeled on the scale of FDR’s Depression-era intervention and the Progressive Era reforms of the early 1900s.

The core argument: advanced AI is already penetrating the economy, next-generation models will help scientists compress decades of discovery into years, and the U.S. has maybe 18 months to build institutions capable of handling what comes next. Altman explicitly warns of a “world-shaking cyberattack” this year as a concrete near-term risk, and flags terrorist groups using AI to develop novel pathogens as a second-tier threat requiring urgent preemption.

What OpenAI Is Actually Proposing

The blueprint has four concrete pillars:

1. A nationally managed AI fund seeded partly by AI companies, distributing economic gains directly to American citizens — closer to an AI dividend than a UBI.
2. Robot taxes on companies replacing workers at scale, with proceeds funding retraining and displaced worker support.
3. An auto-triggering safety net tied to labor market data — if AI-driven unemployment crosses defined thresholds, benefits expand automatically without requiring new legislation.
4. Coordinated government-industry safety protocols for frontier model development, with the federal government as the standards body rather than the models’ developers.

OpenAI frames this as a way to ensure AI’s economic gains don’t concentrate at the top while its disruptions land on everyone else.

The Criticism Is Sharp

The Fortune analysis cuts to the point most concisely: critics argue Altman’s proposals are “regulatory nihilism dressed up as progressivism.” The specific complaint is that OpenAI’s policy agenda focuses on economic redistribution while staying silent on the structural questions that most concern safety researchers — mandatory pre-deployment evaluations, liability for model failures, and limits on capability advancement pace. A fund that pays people while the technology remains ungoverned is a palliative, not a solution.

There’s also a conflict of interest reading: OpenAI is simultaneously the company most likely to build superintelligence and the company writing the social policy around what happens when it does. The instinct to set the terms of one’s own oversight is not unique to tech, but it’s rarely done this openly.

What Actually Changed

Despite the sweeping framing, no legislation is attached to any of this. Altman’s proposals are a discussion document, not a bill. What’s significant is the admission embedded in the piece: that AI is close enough to transformative economic impact that the company needs to get ahead of the backlash. That’s a shift from OpenAI’s earlier posture of “we need to move fast to stay safe.”

The White House has not formally responded. Congressional reaction has been predictably split along existing AI regulation fault lines.

For developers and engineers watching this: the most operationally relevant detail is Altman’s specific warning about AI-enabled cyberattacks arriving in 2026. Whether or not the policy blueprint goes anywhere, that threat model is worth taking seriously in your own infrastructure planning.