White House poised to widen AI safety framework to cover open‑source models

According to recent reporting highlighted by AI policy watchers, the White House is preparing to expand its new AI safety framework so that it no longer focuses solely on proprietary “closed” models from labs like OpenAI and Anthropic. Under the current approach—created via a Trump‑era executive order—the federal government plans to evaluate the most powerful frontier systems for national‑security and cyber risks before they are widely deployed.

The next step, officials say, is likely to bring open‑source models into scope as well. As labs such as Meta and several Chinese companies release increasingly capable systems under permissive licenses, policymakers have grown more concerned that dangerous capabilities could spread in ways that are difficult to monitor or control. A White House official cited in the coverage indicated that future updates to the framework are expected to spell out how large, openly released models should be risk‑assessed and what information developers will be expected to share with the government.

For now, the framework remains an administrative policy rather than binding law, and participation by companies is largely voluntary. That has sparked debate in Washington and the wider AI community over whether the U.S. is moving fast enough—and whether sweeping in open models could chill innovation. For Japanese firms that rely on U.S.‑hosted AI services or plan to market AI‑enabled products in America, these rules could directly affect model choices, compliance strategies, and cross‑border data flows in the coming years.