US AI Model Security Review Excludes Open Source Models
The U.S. government has stated that its cybersecurity review mechanism for advanced AI models currently only applies to closed-source models. Open-source models will not be affected at this time.
The U.S. government has clarified that its cybersecurity review mechanism for frontier artificial intelligence models, announced in June, currently pertains only to closed-source models. Open-source models will at least temporarily not be impacted.
This clarification was made during a closed-door meeting between government officials and leading AI companies. During the meeting, the government defined models requiring voluntary review as closed-source models possessing the most advanced capabilities and posing national security risks. Officials emphasized that the mechanism should not be interpreted as a restriction on already released open-source models.
If an AI company opts to submit a pre-release model for government review, it will undergo a pre-release review period lasting up to 30 days. During this time, multiple government agencies will jointly assess the model, and company employees will not have access to it.
The U.S. government does not expect to publicly disclose the full details of the model review mechanism. However, companies that choose to participate will receive detailed information about the process.