AIMPACT reports that the U.S. ONCD convened a closed-door meeting with OpenAI and others on May 19, 2026, to discuss the proposed signing of an AI executive order, which could be signed by Trump as early as Thursday. The new regulation is a voluntary pre-approval review framework, requiring sharing information with the government before public release, initially set for 90 days, with efforts to shorten it to 14 days before release. Standards for cutting-edge models are jointly defined by the NSA, ONCD, OSTP, and others, with the Department of Defense participating in classified assessments. Previously, Anthropic Mythos's zero-day vulnerability capabilities were seen as a driving factor.

MeNews

2026-05-21 03:40:17

Abstract generation in progress

AIMPACT News, May 21 (UTC+8), according to Beating Monitoring, the U.S. White House National Cyber Director's Office (ONCD) convened closed-door meetings with leading labs such as OpenAI on May 19, 2026, to discuss the upcoming AI executive order. This executive order is expected to be signed by President Trump as early as Thursday. With the new regulation set as a voluntary pre-review framework, the previously feared mandatory licensing concept similar to the Food and Drug Administration (FDA) has been ruled out. The regulation explicitly defines a time parameter, requiring cutting-edge AI models to be shared with the government up to 90 days before public release. However, major labs are lobbying to reduce the sharing window to 14 days before release. The standards for reviewed frontier models are to be jointly defined by agencies such as the National Security Agency (NSA), ONCD, and the Office of Science and Technology Policy (OSTP). These security and intelligence agencies will also collaborate with the Department of Defense to develop a confidential process for evaluating models that meet the criteria. Previously, the zero-day vulnerability discovery capabilities demonstrated by the Mythos model retained by Anthropic directly triggered the acceleration of this new regulation. (Source: BlockBeats)

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

8 Likes

Reward
8
7
9
Share

Comment

Add a comment

GlassFishTankArbitrage

· 10h ago

The Mythos incident is used as a reason, but can the vulnerabilities of closed-source models and the regulatory logic of open-source be conflated?

View OriginalReply0

EraPuzzleMaster

· 10h ago

From 'voluntary' to 'mandatory' often only requires one re-interpretation, and the 90-day buffer period looks like a probationary period.

View OriginalReply0

MirrorPetals

· 10h ago

Reducing from 90 days to 14 days, compressing the time window by six times, with all compliance costs fully passed on to the startup team

View OriginalReply0

WatercolorInAGlassBottle

· 10h ago

The Anthropic vulnerability became the catalyst; regulatory narratives always require a 'crisis' to trigger them.

View OriginalReply0

Can'tSleepWithoutSigningThe

· 11h ago

If Trump signs it, this will be the first brick of the MAGA version of AI policy.

View OriginalReply0

OtcMoonwalker

· 11h ago

NSA + Department of Defense intervention assessment, is this framework truly secure or just control? The boundaries are becoming increasingly blurred.

View OriginalReply0

0xLateComer

· 11h ago

Voluntary review sounds gentle, but a 14-day waiting period is a fatal blow to iteration speed. Can OpenAI accept this?

View OriginalReply0