U.S. lawmakers demand early government access to OpenAI’s next model

Silicon Valley usually tests AI models behind closed doors before hitting deploy. Now, Washington wants to be in the testing room first.

A group of Senate Democrats and one independent lawmaker just sent a formal letter to OpenAI CEO Sam Altman, pressing the startup on its internal safety standards and treatment of whistleblowers.

A 12-question grilling

The letter, first obtained by The Washington Post, lays out 12 pointed questions for Altman. The most striking one is Item 9: Will OpenAI commit to handing over its next foundation model to U.S. government agencies for pre-deployment testing, review, analysis, and assessment?

Lawmakers are also demanding hard numbers. They asked what percentage of computing power OpenAI currently dedicates to AI safety research, pushing for a firm commitment of at least 20%. The letter also presses the company on protocols to stop foreign adversaries or malicious actors from stealing its intellectual property.

The congressional push was triggered by whistleblower reports alleging OpenAI relaxed safety evaluations for GPT-4 Omni just to avoid missing its launch target.

Those whistleblowers also claimed that raising safety concerns internally resulted in retaliation and allegedly illegal non-disclosure agreements, leading them to file a formal complaint with the U.S. Securities and Exchange Commission in June 2024.

The alarm bells inside the company run deep. Former OpenAI employee William Saunders recently revealed he quit because he believed the research trajectory posed an existential threat to humanity, likening OpenAI to the Titanic in 1912. Saunders clarified that while current ChatGPT models are fine, advanced superhuman intelligence in the pipeline is what keeps him up at night.

Big Tech allies are already backing away from the blast radius. In July, both Microsoft—despite its $13 billion investment—and Apple gave up their seats on OpenAI's board as antitrust and regulatory scrutiny intensified.

Why it matters

For developers and everyday users, this will not break your current workflows, but it could dramatically slow down the arrival of OpenAI's next flagship models. If government sign-off becomes a mandatory checkpoint, the breakneck pace of generative AI releases is going to hit a wall.

For the AI industry, this marks a fundamental shift from post-launch oversight to pre-market gatekeeping. If OpenAI opens its next model to federal evaluation before release, competitors like Anthropic and Google will face intense pressure to surrender the same access, establishing a new de facto compliance standard for frontier labs.

Lawmakers were clever to focus on the 20% compute commitment and whistleblower protections, two sore spots where OpenAI has recently struggled to defend its reputation. However, we think this demand exposes a glaring practical issue: U.S. regulatory bodies currently lack the compute infrastructure and specialized talent required to properly stress-test frontier AI models. Without clear benchmarks, pre-deployment review looks like a recipe for bureaucratic gridlock.