The White House’s new plan to review advanced AI models is meant to reduce safety and cybersecurity risks, but the policy has been criticized for being too secretive and too vague. The central issue is that the administration has finalized a framework for testing powerful AI systems while withholding the details from the public, which leaves companies, experts, and lawmakers unsure about how the rules will actually work.
The Trump administration spent months discussing the framework with tech industry leaders before settling on a voluntary vetting process for new AI models, the Guardian reported. Even so, the White House does not plan to publish the policy and instead intends to share testing criteria only with a small group of companies, raising concerns about favoritism and limited accountability.
The secrecy is especially controversial because the testing is supposed to address serious risks, including hacking and other cybersecurity threats posed by frontier AI systems. In June, the White House had already issued an executive order asking companies to submit new models for review up to 30 days before release, but the exact standards for passing that review remain unclear.
Another concern is the narrow scope of the framework. Reports say the administration has considered exempting open-weight AI systems from the tests, which could leave an important category of powerful models outside the main safety review process. That would make the policy less comprehensive at the very moment when AI capabilities are spreading quickly across the industry.
Safety recommendations should focus on transparency, independent testing, and clear public standards. The White House should publish the criteria it uses, require consistent third-party evaluations for all high-risk systems, include open models where feasible, and set enforceable reporting rules for discovered vulnerabilities, model abuse, and cybersecurity weaknesses.
Read the original article: