Breaking

US Government Asks OpenAI to Stagger GPT-5.6 Release Over Security

June 25, 2026 at 18:46 EDT

Policy & Regulation
Foundation Models
Security

The Trump administration has asked OpenAI to limit the release of its next frontier model, "GPT-5.6," citing national security concerns including cybersecurity. The model is expected to start as a limited preview for government-approved partners.

June 25, 2026 · AI Policy

US Government Asks OpenAI to Hold Back GPT-5.6 Over Cyber Fears

In a first, federal cyber and science offices requested that OpenAI's next model launch only as a limited preview — with access granted one approved customer at a time, pending a security review.

1st

time the US government has asked a US AI company to restrict a model before release

1×1

access approved customer by customer in an unusual process

Limited

preview only, with hopes for wider release within a few weeks

CyberGym benchmark — vulnerability reproduction

The capability edge that triggered the caution (% score)

85.6%

GPT-5.5-Cyber

83.8%

Mythos 5

ExploitGym

39.5%

SEC-bench Pro

69.8%

Why the brakes went on

GPT-5.6 was judged to have "Mythos-like" cybersecurity capabilities — raising the risk of advanced models leaking to bad actors.

Nation-state actors

Cybercriminals

Malicious insiders

From capability to controlled release

Trump signs AI security EO — voluntary testing framework

→

GPT-5.6 flagged with "Mythos-like" cyber capability

→

Limited preview + per-customer security review

Supportive view

Some welcome the government's swift response. GPT-5.6 is offered mainly for defensive use to verified security teams.

Critical view

Others worry about reliability, real-world effectiveness, and the impact on global competition with Chinese open-source models.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.