White House Issues 90-Minute Deadline for AI Developers to Disclose Test Data

Federal Oversight and the June 12 Ultimatum

The June 12 directive originated from the White House Office of Science and Technology Policy (OSTP). It required firms to submit internal security audits and red-teaming data for any models currently in training that exceed a specific compute threshold. This mandate builds upon the broader regulatory architecture established by the October 2023 Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, which first formalized the requirement for developers to share safety test results with the federal government.

According to a spokesperson for the National Security Council, the move was designed to ensure that national security implications are addressed before public deployment. This deadline forced a rapid shift in operational priorities for Silicon Valley executives who had previously argued that voluntary commitments were sufficient for managing safety risks. The White House justified the urgency by citing the emergence of new “dual-use” capabilities in frontier models—systems capable of assisting in cyberattacks or the development of biological agents—which necessitate, in the administration’s view, instantaneous federal visibility.

Silicon Valley’s Response to Regulatory Pressure

Industry leaders reacted with caution to the sudden timeline. Representatives from several major labs held an emergency briefing with administration officials on the afternoon of June 13 to discuss the feasibility of the reporting requirements. This meeting included senior policy leads from the major AI firms, who sought to clarify whether the “90-minute” window applied to the transmission of raw data or summarized executive reports.

We are committed to transparency, but the 90-minute window for a request of this complexity ignores the technical reality of how model safety data is aggregated and verified across decentralized engineering teams.

The industry’s pushback centers on the concern that such rapid mandates could stifle innovation or expose proprietary model architectures to premature public scrutiny. While companies like Microsoft and Meta have historically aligned with government safety frameworks—such as participating in the U.S. AI Safety Institute’s voluntary testing programs—the current demand for granular, real-time safety data represents a departure from the collaborative, slower-paced regulatory approach seen throughout 2025. During that period, the focus remained on the “AI Safety Summit” model, which emphasized international cooperation and non-binding commitments between the government and private sector labs.

Contrasting Regulatory Frameworks

The current tension highlights a widening gap between the administration’s focus on immediate risk mitigation and the industry’s preference for long-term, standardized safety protocols. The disagreement reflects a fundamental friction regarding how “safety” is defined when models are still in the training phase.

• The White House (June 2026): Focuses on mandatory, time-bound disclosures for compute-heavy models, prioritizing national security and preventing model misuse through direct federal intervention.
• Industry Consortiums (2025-2026): Prioritize voluntary, peer-reviewed standards that emphasize model performance benchmarks and gradual release cycles, often managed through independent bodies rather than direct government mandates.

Market analysts note that the suddenness of the White House request has unsettled investors. On the morning of June 15, 2026, shares in companies heavily exposed to AI infrastructure saw minor volatility as traders weighed the potential for future compliance costs. In the context of broader sector performance, this reaction mirrors the market sensitivity observed during the initial introduction of the EU AI Act’s compliance requirements, where investors reacted negatively to the prospect of increased administrative overhead for frontier model developers.

Implications for Future AI Development

The conflict over the June 12 deadline is likely to influence the upcoming legislative session in Congress. Several senators have signaled their intent to codify these reporting requirements into law, which would effectively end the era of voluntary safety compliance. This proposed legislative shift follows a series of hearings held by the Senate Judiciary Subcommittee on Privacy, Technology, and the Law, where members have debated the necessity of a federal licensing regime for large-scale AI training runs.

If the administration continues to use short-term deadlines to force compliance, industry experts suggest that companies may move sensitive research operations to jurisdictions with more flexible oversight, a phenomenon known as “regulatory arbitrage.” This concern was previously documented in industry white papers regarding the global distribution of GPU clusters. Conversely, proponents of the current mandate argue that the potential for catastrophic failure in unmonitored models necessitates a more aggressive posture from federal regulators, drawing parallels to the established oversight mechanisms used by the Nuclear Regulatory Commission or the Federal Aviation Administration.

As of June 15, 2026, the White House has not stated whether it will levy penalties against firms that failed to meet the initial 90-minute filing requirement, leaving the status of the ongoing standoff uncertain. The lack of clarity regarding enforcement mechanisms—whether through administrative fines or litigation—continues to be a primary point of concern for corporate legal teams currently reviewing their obligations under the June 12 order.

Find more reporting in our Business section.