Trump Administration Drafts AI Safety Framework for Closed-Source Models

The White House is preparing a new voluntary

The White House is preparing a new voluntary review system aimed at assessing national security risks tied to advanced artificial intelligence, according to people familiar with the matter. The initiative, still taking shape, would apply specifically to closed-source AI systems—those that do not disclose their underlying code—while leaving open-source models outside its scope.

Officials describe the effort as a light-touch approach that encourages companies to submit their most powerful AI technologies for federal review, rather than imposing mandatory rules. The framework is expected to focus on potential misuse in areas like cyberattacks, biological threats, or critical infrastructure vulnerabilities. It marks an early step by the Trump administration to shape AI policy without triggering industry backlash over heavy-handed regulation.

Under the plan, participation would be voluntary, but the administration hopes that offering a clear federal approval process will attract major developers, given the growing pressure from Congress and state lawmakers to address AI dangers. The exclusion of open-source models—where code is publicly available—reflects a recognition that such systems are harder to control and that forcing compliance could push development overseas.

The framework is still being finalized, and details

The framework is still being finalized, and details could shift before an official announcement. Administration aides have stressed that the goal is to build trust with the tech sector while ensuring that the most consequential AI tools receive a security once-over. A formal rollout is expected in the coming weeks, though no firm date has been set.

Industry analysts note that the voluntary nature of the plan may limit its impact, but they see it as a pragmatic starting point. Companies that opt in would receive a federal seal of approval, potentially easing concerns from investors and foreign buyers. Meanwhile, critics argue that leaving open-source AI entirely unexamined could create a blind spot, especially as such models become more capable and widely adopted.

For now, the White House is proceeding cautiously, aiming to balance innovation with oversight. The framework is likely to be a centerpiece of the administration’s broader tech agenda, signaling that AI security remains a top priority—even if the approach is softer than some lawmakers have demanded.