Two lobbying efforts converge on Washington as Congress weighs new AI oversight powers
Nvidia CEO Jensen Huang, Meta, and Microsoft signed an open letter this week defending open-weight AI models, the kind whose internal parameters are publicly released and can be run, modified, or redistributed without a license. The letter, published July 24, argues that open models strengthen safety and cybersecurity, and urges US policymakers to avoid broad restrictions that would concentrate AI power among a small number of closed-model providers.
A bipartisan House bill, the AI Kill Switch Act, would grant the Department of Homeland Security authority to halt AI models deemed dangerous, according to Politico, which viewed the legislation. A separate letter from the Little Tech Association, a coalition of nearly 200 US startups, makes a parallel argument: the government should not ban Chinese open-weight AI models outright, but instead pursue targeted security safeguards.
Two coalitions, one pressure point
The two efforts share a conclusion but not a constituency. The Big Tech letter, signed by some of the largest closed-model developers in the world, frames open-weight AI as a public good and a competitive asset for the United States. The startup coalition focuses on access. Smaller companies have built products on top of Chinese open-weight models, and a blanket ban would disrupt those businesses directly.
The Little Tech Association letter also surfaces an intellectual property dispute running beneath the access debate. The coalition acknowledges allegations that some foreign models improperly distilled training data from US competitors, a claim that remains disputed and unresolved. The startups argue that IP concerns and national security concerns require different policy responses, and that conflating them risks overreach.
What the legislation actually proposes
The AI Kill Switch Act would give DHS authority to shut down AI models following what the bill describes as dangerous autonomous AI incidents. The bill does not appear, based on available sourcing, to target open-weight models specifically. But the letter's authors anticipate that any broad federal authority over AI deployment could be applied to publicly released weights in ways that would effectively remove them from circulation.
New York's RAISE Act adds a state-level dimension. That legislation would impose registration, disclosure, and safety-planning requirements on major AI developers, moving ahead of federal action. Whether major AI developers encompasses open-weight model distributors or only closed-model API providers is a compliance question the bill's drafters have not yet answered publicly.
The competitive logic underneath the safety argument
Meta is the most prominent open-weight model developer among the signatories, having released successive versions of its Llama model family under permissive licenses. For Meta, open-weight distribution is a business choice: it builds an ecosystem of developers dependent on Meta's model architecture rather than a rival's. Microsoft and Nvidia benefit differently. More models running on more hardware and cloud infrastructure means more demand for their products regardless of which model wins.
That alignment of commercial interest with the pro-openness argument does not make the argument wrong. It does mean the safety framing deserves examination on its merits, separate from the business case. The coalition argues open models let independent researchers audit for vulnerabilities that closed-model providers might not disclose. Critics of open-weight release argue that public weights lower the barrier for misuse in ways that post-hoc auditing cannot address. Neither letter resolves that tension; each asserts one side of it.
The next test will come when the AI Kill Switch Act advances to committee markup, when the bill's language around model types and DHS authority faces its first formal scrutiny.
Get weekly leaderboard changes and AI video model launches delivered to your inbox.