OpenAI and Anthropic Join Forces to Lobby for Restrictions on Open-Source AI Models
Alina Collins
OpenAI and Anthropic — fierce commercial rivals — are jointly pressing Washington to flag the safety risks of open-source AI models. The outcome of this lobbying push could determine who gets to release powerful AI going forward.
They compete on products — so why team up on policy?
OpenAI and Anthropic fight head-to-head in the market, but have found common ground on policy lobbying: warning U.S. policymakers about the risks of open-weight AI models — models whose core parameters are published for anyone to download and use.
Anthropic CEO Dario Amodei's core argument: once an open-weight model is released, the developer cannot revoke access, update safety guardrails, or prevent misuse.
This means → two closed-source giants, despite being rivals, share a strong interest in shaping the rules around who gets to release powerful models.
Is this a safety concern — or regulatory capture?
Open-source advocates see the inability of any single company to control usage as a feature, not a bug.
Trump administration adviser David Sacks called the lobbying effort "regulatory capture" outright: rules written in the name of safety that, in practice, raise the bar for competitors to release models.
In plain terms = if releasing an AI model requires clearing high regulatory hurdles, the companies best positioned to benefit are those that already have the resources and compliance teams — the rule itself becomes a moat.
How did "distillation" and Chinese models get pulled in?
OpenAI has warned about distillation — using one company's model to train another's. U.S. Trade Representative Jamieson Greer went further, framing China's distillation practices as intellectual-property theft.
Treasury Secretary Scott Bessent said this week that Chinese models must face the same scrutiny as American ones, with an investigation to begin within weeks.
Advocates warn that labeling distillation as theft could end up restricting all open-weight models — including those developed independently, with no connection to China. This reflects a recurring pattern: once a safety narrative merges with geopolitics, the blast radius easily exceeds the original target.
Does the U.S. even have safety standards of its own?
Not yet. The U.S. has no detailed safety standards for domestic AI models; the framework is still being drafted.
An OpenAI spokesperson said China's open-weight progress "is not an argument against openness," but called for a "coherent national framework" to evaluate new models and manage risk.
This means → OpenAI says it is not against openness while pushing for a review system it is best equipped to pass — and the framework's specifics remain undisclosed.
Are the two companies really in lockstep?
No. At the state level, clear differences have surfaced. Both back a Massachusetts frontier-AI safety bill — but they back different versions.
The version Anthropic endorsed requires independent model testing every six months with public reporting, and empowers the state attorney general to sue and impose civil penalties. The version OpenAI endorsed includes none of those provisions.
In plain terms = Anthropic is willing to accept stricter outside oversight; OpenAI prefers a lighter framework. Behind the joint lobbying, the two companies draw the line on "how strict" in very different places.
What to watch next?
OpenAI CEO Sam Altman is expected to meet with lawmakers and the White House next week to discuss plans for releasing new models.
Former Biden administration tech adviser Suresh Venkatasubramanian warned that cutting off access to Chinese models entirely would be "a major problem for doing research", since open-weight models are especially valuable to scientists.
This means → the central question in this policy fight is whether two closed-source giants can turn a safety narrative into binding regulation — and if they succeed, the space for open-source AI could shrink dramatically.
Content is for reference only, not financial advice.