Tech giants lobby to block regulation of open-weight AI models
Nvidia, Microsoft, Meta, Palantir, and 20+ companies warn the White House against 'premature restrictions' on open-weight AI models, framing their letter as a plea for innovation but effectively opposing binding safety standards.
A coalition of the largest AI developers—Nvidia, Microsoft, Meta, Palantir, and over 20 others—has sent a letter to the White House urging against what they call 'premature restrictions' on open-weight AI models. The letter, reported by CNBC, argues that limiting how models are released would harm U.S. competitiveness and ‘innovation.’ This is a direct lobbying push aimed at preserving the industry’s current free-releasing model, which allows powerful, unvetted AI systems to be downloaded, modified, and deployed with minimal oversight.
Open-weight models—AI systems already trained and whose underlying weights (the core learned parameters) are made public—pose unique risks. They can be fine-tuned to remove safety guardrails, weaponized for disinformation, or used in surveillance and autonomous systems without any federal check. The signatories' letter arrives as the Trump administration considers new executive orders or agency rules on AI safety, following a year of voluntary pledges that have proven insufficient: no major incident has triggered binding action, and the federal AI safety testing framework remains non-binding.
The letter’s call to avoid 'premature restrictions' is a known industry tactic: stall regulation until the technology is so entrenched that any later rule is harder to enforce. The concrete outcome would be continued release of high-risk models with no mandatory safety evaluation, no public transparency on training data or testing results, and no binding requirement to keep human oversight in critical applications. The harm is systemic: without federal guardrails, the costs of AI failures—from algorithmic bias in hiring to autonomous weapons drift—are externalized onto the public, while the companies capture the profits.
The White House should reject this letter's framing. The alternative is not a ban on open-weight models but a tiered approach: mandatory safety assessments before release for models exceeding certain compute or capability thresholds (like the EU AI Act's tiered framework), public disclosure of testing results, and liability for demonstrably foreseeable harms. Several of the signatories—especially Meta and Microsoft—already operate in jurisdictions like the EU that impose such rules; they are arguing for less protection in the U.S., not against all regulation.
The humanitarian alternative
Congress and the White House should adopt a tiered regulatory framework for open-weight AI models, modeled on the EU AI Act: models above a compute or capability threshold must undergo mandatory safety evaluations by an independent federal body (e.g., NIST), with results published before release. Lower-risk models would be exempt. This approach preserves openness for small-scale innovation while ensuring that powerful, dual-use systems cannot be deployed without accountability. The National AI Safety Institute should be given binding enforcement authority, and the FTC should be empowered to act against demonstrable harms from unreleased or under-tested models.
Falsifiable predictions
What this entry claims will happen, and what data would prove it wrong. The Reckoner revisits these against current reality.
- Within 12 months, the White House will issue an executive order that requires safety testing only for the largest frontier models (exceeding 10^26 FLOPs), leaving open-weight models below that threshold unregulated.
- Three of the five signatory companies (likely Meta, Nvidia, and Palantir) will continue releasing new frontier open-weight models without mandatory safety evaluations within 6 months of the letter.
- The industry coalition will mount a lobbying campaign against any bill that includes a mandatory safety-assessment requirement for models over 10^22 FLOPs, spending at least $10 million on ads and Hill meetings by end of 2026.
Original source — excerpted
news Nvidia, Microsoft, Meta warn against 'premature restrictions' of open-weight models"The White House is visited by tourists on July 17, 2025 in Washington, DC. Nvidia , Microsoft , Meta , Palantir and more than 20 other companies released a let..."