US AI security review may be expanded to open source models: cutting-edge definition and benchmarking GPT-5

U.S. AI security review may be expanded to open source models: Frontier Definition Benchmark GPT-5.6, voluntary framework faces tightening pressure According to "Wired" magazine citing sources, the U.S. cutting-edge artificial intelligence model cybersecurity review mechanism, which currently only targets closed-source models, will "almost certainly" be expanded to open source (open weight) models in the coming months. The definition of "frontier" refers to models with equivalent capabilities to Anthropic Claude Mythos or OpenAI GPT-5.6, but the report did not give a more detailed description of the corresponding version. According to people familiar with the matter, this review framework is still voluntary. Some U.S. officials are aware that mandatory scrutiny could stifle AI development and are reluctant to upgrade it to mandatory compliance requirements for now. However, the signal that the scope of review has expanded from closed source to open source has been very clear, which means that manufacturers such as DeepSeek and Meta that take the open source route may also need to face cybersecurity scrutiny from the US government in the future. The open source model faces a "market squeeze" effect. The ripple effects of expanded scrutiny extend beyond compliance itself. When approved closed-source models receive official endorsement, U.S. companies may reduce their interest in using unapproved open source models, thereby inhibiting the willingness of local manufacturers to launch open source models. This is also one of the factors the White House considers when making decisions. In other words, even if the review is nominally voluntary, the "soft constraints" at the market level are enough to compress the living space of the open source model. At the same time, there have been calls within the U.S. government to tighten the regulatory network and promote the gradual transition of voluntary review into mandatory requirements. The core of the controversy surrounding the open source model is that open weight allows anyone to download and deploy the model, making security risks more difficult to control. However, excessive tightening may stifle the innovative vitality of the open source ecosystem and push developers to overseas markets with looser regulations. How to strike a balance between security and open innovation will be a key aspect of the direction of U.S. AI policy in the coming months. The impact of this decision will go far beyond the borders of the United States and directly reshape the open-source and closed-source competition of large global models. via AI News (author: AI Base)