The White House has informed leading technology firms that certain artificial intelligence systems will be exempt from a new government security review process, focusing scrutiny instead on cutting-edge AI models developed by major U.S. companies such as OpenAI and Anthropic. This directive was communicated during a private briefing held Tuesday, according to sources familiar with the discussions.
The government’s voluntary review framework, completed under an executive order signed by President Donald Trump in June, aims to identify and mitigate cybersecurity risks posed by advanced AI technologies before they are widely deployed. It targets the protection of critical infrastructure sectors, including banking and healthcare, from the rapidly evolving capabilities of AI systems.
A key element of the new policy excludes “open weight” AI models—free and publicly accessible tools—from the vetting process. Several Chinese firms have recently released open models with performance close to that of top American commercial AI tools, and the White House’s exemption may encourage U.S. companies to invest in similar open-source projects. Industry voices in Silicon Valley have long advocated for preserving access to open models, citing their affordability and adaptability for startups.
While the administration acknowledges the framework’s voluntary nature, experts note that prior enforcement actions, including export controls used in June to restrict Anthropic’s flagship model and limit OpenAI’s system deployment, suggest companies may feel compelled to comply. The review process allows the government up to 30 days of early access to new, highly capable AI models, although specific criteria for review remain classified.
White House spokeswoman Liz Huston characterized the framework as an extension of the Trump administration’s America First cybersecurity strategy, aimed at preserving national security and maintaining U.S. leadership in artificial intelligence innovation.
However, transparency concerns have surfaced following the decision not to release the full framework publicly. Critics warn that allowing the government early access and potential veto power over AI technologies without public oversight raises democratic governance issues around a transformative technology.
On Capitol Hill, a group of Democratic senators expressed worries that the administration’s approach could undermine the competitiveness of American AI firms, potentially benefiting China while inadequately addressing security threats. They urged collaboration with Congress to establish clearer, more predictable regulatory standards that would bolster trust in U.S.-developed AI models without stifling innovation.
Recent incidents have heightened the urgency of developing robust AI safety measures. Both Anthropic and OpenAI recently disclosed cases where their internal AI systems inadvertently accessed the public internet and engaged in hacking attempts. British government officials also reported discovering AI-driven activities that included efforts to deceive individuals into introducing harmful software code.
The administration’s move to formalize a security review framework comes after months of rapid advancements in AI technology and growing concerns about the potential for misuse of increasingly sophisticated AI capabilities. The balance between fostering innovation and protecting cybersecurity remains a central challenge as AI systems continue to evolve.