OpenAI pitches its own AI policy blueprint to Washington

OpenAI just handed Washington a custom-made playbook for how it wants to be regulated.

On Monday, the company published an “economic blueprint” for AI, framing massive government spending and streamlined oversight as matters of urgent national security. With a forward from VP of global affairs Chris Lehane, the document lays out exactly what OpenAI wants from the incoming Trump administration and Congress: billions for infrastructure, an end to state-by-state rules, and a free pass to train on copyrighted data.

Pave the grid, scrap the state red tape

OpenAI’s biggest operational headache right now isn't just compute—it's power and paperwork.

In 2024 alone, state lawmakers floated nearly 700 AI-related bills, creating a patchwork of conflicting rules. OpenAI singled out measures like Texas’ Responsible AI Governance Act for putting heavy liability on open-source developers, having previously lobbied hard against California’s SB 1047.

CEO Sam Altman recently told Bloomberg that existing federal efforts like the CHIPS Act haven’t been as effective as hoped. He openly aligned himself with Donald Trump’s frustration over how hard it is to build data centers and power plants in the U.S. due to bureaucratic cruft.

To keep the AI buildout on track, the blueprint urges the federal government to dramatically boost spending on transmission lines and green-light energy sources like nuclear, solar, and wind. Rivals like Meta and AWS have already hit roadblocks trying to secure nuclear power, and OpenAI wants Washington to clear the runway.

Voluntary guardrails and open-season data

When it comes to safety rules, OpenAI wants Washington to keep things strictly optional.

The blueprint advocates for a “voluntary pathway” where developers collaborate with national security agencies on model evaluations and threat briefings, rather than facing mandatory federal mandates. On the international front, it pushes to export frontier models to allies to lock them into U.S. tech ecosystems while tightening export controls against foreign rivals.

Then comes the copyright fight. Amid a barrage of author and publisher lawsuits, OpenAI argues that models should be legally allowed to learn from publicly available information, including copyrighted works, just like humans do.

OpenAI's defense is straight geopolitical leverage: if the U.S. restricts training on public data, competitors overseas will simply train on it anyway, handing the economic upside to rival nations.

To make sure this message lands, OpenAI has been aggressively staffing up its Washington bench. The company brought in former NSA chief Paul Nakasone, ex-Defense official Sasha Baker, and former Commerce chief economist Aaron Chatterji. Its lobbying tab also hit $800,000 in just the first half of last year, more than triple its total spend for 2023.

Why it matters

We think this blueprint is less of a policy paper and more of a corporate wishlist dressed up in the American flag.

For enterprise users and developers, federal preemption of state laws would bring much-needed clarity, replacing 50 different rulebooks with a single standard. But the voluntary approach OpenAI is pitching looks like regulatory capture in plain sight. It leaves safety enforcement entirely to private goodwill while shifting the massive capital costs of power grids and nuclear plants onto public shoulders.

By tying copyright exceptions and nuclear permits directly to national security, OpenAI is daring policymakers to say no. If Washington adopts this framework, it will entrench a handful of well-connected frontier labs while making it far harder for independent watchdogs and smaller creators to hold them accountable.