White House Eyes Open Models For Safety Rules. About Time. Regulators Still Don't Understand What They're Regulating.
White House officials are preparing to revise the Trump administration's AI guidelines to include open models in a federal safety framework. The current framework requires the most powerful models from US labs to undergo government safety testing before public release. Sources tell WIRED that open models, which are freely downloadable and modifiable, may soon fall under similar oversight.
This illustrates the Cat and Mouse Problem of regulation. Policymakers build frameworks around what they can measure, not what matters. Open models, by definition, escape the controlled-release paradigm that closed-model regulation relies on. Once weights are public, safety testing after the fact is theater. The reader should understand that regulatory frameworks always lag behind the actual distribution mechanism of the technology they target.
The White House is leading this effort, revising AI guidelines developed under the Trump administration. The framework currently covers the most powerful models created by US labs, per reporting from WIRED sources familiar with the matter.
- Visit huggingface.co and search for an open model like 'Llama' or 'Mistral' to see how freely downloadable AI weights are distributed. Expected outcome: You will find model pages with download buttons and no gatekeeping.
- Read the model card on any open model page, which lists capabilities and limitations. Expected outcome: You will notice there is no government safety certification, only self-reported information from the creators.
- Reflect on how you would regulate something that anyone can download, modify, and redistribute with zero oversight. Expected outcome: You will understand why adding open models to a testing framework is far more complex than it sounds.