U.S. Plans to Expand AI Cybersecurity Review to Open-Source Models

Recently, the U.S. cybersecurity review mechanism for closed-source AI models could expand to cover open-source models within the next few months. The review defines “frontier” model capabilities as comparable to Anthropic’s Claude, Mythos, or OpenAI’s GPT-5.6. Currently voluntary, the review framework has raised concerns among some officials that mandatory reviews might stifle AI innovation. U.S. companies, after securing approval for their closed-source models, may reduce their use of unreviewed open-source models, potentially dampening corporate willingness to release open-source models. Calls for stronger oversight are growing within the U.S. government, with some officials arguing that as the review process for closed-source models matures, businesses will increasingly favor reviewed closed-source models and rely less on open-source alternatives. The White House is weighing the impact of this review mechanism on AI development and considering how best to balance security with innovation.