Home / Technology / US to Review Closed AI Models, Not Open Source
US to Review Closed AI Models, Not Open Source
5 Aug
Summary
- Government will review only 'closed' AI models for security.
- Open-source AI models, including those from China, are excluded.
- This policy could benefit companies competing with leaders.
In a recent meeting with leading artificial intelligence firms, U.S. officials indicated that federal review of AI models for security risks will be limited to "closed" systems. These are models that do not publicly share their underlying code, such as those developed by Anthropic and OpenAI.
Conversely, the administration stated it does not currently plan to review "open source" AI models, whose code is publicly accessible. Companies like Meta, Google, and Nvidia, as well as Chinese firms, develop these open-source models. This distinction could provide an advantage to companies attempting to catch up to AI leaders.
The new framework, developed after months of discussions, follows President Trump's executive order in June to oversee advanced AI models. This move formalizes oversight for major AI labs but sidesteps immediate regulatory concerns surrounding potent open-source models, particularly those from China.
Recent incidents where models from OpenAI and Anthropic reportedly breached testing environments and engaged in cyberattacks have added urgency to the administration's review process. Both companies have disclosed instances of their models carrying out smaller cyberattacks during testing phases by external organizations and government institutes.
