President Signs Voluntary AI Pre-Release Sharing Order

According to The Atlantic, President Donald Trump signed an executive order that creates a process for leading AI firms to voluntarily share certain upcoming models with the federal government for safety testing as much as 30 days before broader release. The Atlantic reports the order also asks firms such as OpenAI and Anthropic to help strengthen federal, state, and local cyber defenses. White House spokesperson Liz Huston is quoted calling the policy a "common-sense approach of collaborating with industry to balance innovation and security," and Daniel Remler of the Center for a New American Security told The Atlantic the rule "effectively formalizes what has already been happening between the US government and the leading AI companies." The Atlantic also reports the president had canceled a planned signing hours before the ceremony, then later issued the order. CNBC and NPR report the final review window was set at about 30 days, shortened from up to 90 days in an earlier draft.

The reported arrangement is a voluntary, pre-release disclosure window for safety testing. Editorial analysis - industry-pattern observation: voluntary, time-limited model-access arrangements are commonly framed as a compromise between rapid deployment and external auditing, but they vary in scope, including what artifacts are shared and whether weights or only evaluation endpoints are provided. Where governments lack statutory authority to compel access, such frameworks tend to rely on existing relationships, nonbinding memoranda, or contractual agreements between agencies and vendors.

Reporting from The Washington Post documents a December 2025 executive order that directed the federal government to pursue lawsuits against states whose AI regulations were judged to undermine "global AI dominance," and R Street commentary describes related actions emphasizing federal preemption and reduced regulatory constraints. A press release from Representative Zoe Lofgren criticized the administration's approach as unlawful and urged congressional engagement. Taken together, the coverage shows the administration pairing an industry-collaboration framing for federal oversight with measures intended to limit state-level regulation.

• •Whether major model developers accept the voluntary sharing window and what artifacts they provide.
• •Whether the Department of Justice follows through on directives discussed in commentary; R Street notes the related order directs the Attorney General to create an AI Litigation Task Force to challenge conflicting state laws.
• •Any congressional action to create binding standards or codify disclosure obligations.

For practitioners, the immediate operational change is likely modest if, as Remler told The Atlantic, the order mainly formalizes preexisting arrangements. Industry-pattern observation: when governments formalize voluntary review channels without new enforcement tools, organizations running models must weigh legal risk, customer expectations, and the cost of preparing models for external testing. Legal and compliance teams should also track state-federal conflicts, since reporting indicates continued federal efforts to challenge state AI rules.

The order creates a voluntary, short-window process for federal safety testing and sits alongside prior actions aimed at preempting state regulation. Coverage frames the move as more procedural than transformational, while litigation over state laws remains an active vector that will shape how the policy affects deployments and compliance.