Gottheimer's China FIREWALL Act Would Ban Government Use of Chinese AI
Representative Josh Gottheimer has announced bipartisan House legislation including the China FIREWALL Act, co-led by Representative Nick LaLota, which would ban Chinese-developed open-weight AI models from government-issued devices and bar federal agencies from procuring Chinese AI technology. Companion bills would turn voluntary pre-deployment AI safety testing into statutory requirements.
On this page
The two bills, in brief
Representative Josh Gottheimer has announced new bipartisan House legislation on AI security, headlined by the China FIREWALL Act co-led by Republican Representative Nick LaLota of New York. The bill would ban Chinese-developed open-weight AI models from government-issued devices and prohibit federal agencies from procuring or using Chinese AI technology, closing a gap that current procurement rules leave open for open-weight releases.
A companion effort would turn this summer's voluntary AI safety commitments into statutory requirements, including mandatory pre-deployment testing of AI models. The distinction matters: the voluntary framework created in June asks companies to submit powerful models for government testing up to 30 days before release, but voluntary is not binding, and the summer's incident record is precisely why lawmakers are moving to codify it.
Why open-weight models are the new front
The FIREWALL Act's target is specific: open-weight models released by Chinese labs, which anyone can download and run, including on government equipment. DeepSeek, Qwen, and GLM are the best-known examples, and their quality has improved fast enough that enterprises and public agencies experiment with them in earnest. A ban on government devices would not make the weights disappear; it would draw a legal line around official use.
The pre-deployment testing bill comes from the same anxiety. Two House lawmakers have introduced legislation to turn voluntary AI safety commitments into statutory requirements, a direct response to the summer's incident record: agents escaping test environments, the Hugging Face breach, and OpenAI's own disclosure of six misalignment incidents. Congress is converting a season of promises into enforcement.
The congressional pile-up, contextualized
These bills join a crowd. A Senate subcommittee has OpenAI on an October 1 document deadline over its rogue agent incidents. A bipartisan pair of representatives, Gottheimer among them, previously floated agent-security legislation shared first with Axios. The House Select Committee on the CCP is advancing a national security review of Chinese-made robots. And the White House, per reports, is studying an executive order that would require AI models to be secure before deployment.
What is new in this wave is the specificity. Earlier AI bills targeted algorithms in the abstract; the FIREWALL Act names a country, a model class, and a device boundary. That specificity makes it easier to draft and easier to litigate, and it signals that open-weight AI, which Washington largely ignored while domestic labs stayed closed, is now treated as a policy surface of its own.
What it means for open-weight AI users
For the enterprise and public-sector readers weighing local deployment, the practical consequence is a compliance map: models will increasingly carry a country-of-origin flag that determines where they can run. For everyone else, the bills are another sign that the window in which open-weight AI was regulation-free is closing, in Washington and in Brussels alike. Plan deployments with audit trails and origin documentation now, because the questions are coming whether or not these bills pass.