White House Asks OpenAI to Slow GPT 5.6 Release Over Safety Concerns

OpenAI’s upcoming release of its newest model, GPT 5.6, will follow an unusual path. Instead of a broad public launch, the company plans to initially share the model only with a select group of partners, following direct requests from the Trump administration, according to a report from The Information.

Government Oversight Shapes Release Plans

At a meeting this week, OpenAI CEO Sam Altman reportedly told staff that the government would be “approving access customer by customer” during an initial preview period. Altman added that if the limited rollout proceeds smoothly, the company hopes to follow with a broader release a “couple of weeks later.” This marks a notable shift for OpenAI, which has historically released models widely and quickly.

The agencies involved in the request include the Office of the National Cyber Director and the Office of Science and Technology Policy. According to The Information, OpenAI staffers “worked closely” with these government bodies on the upcoming release. The Trump administration, which earlier positioned itself as favoring a “hands off” approach to AI, has recently moved toward more federal oversight. Earlier this month, Trump signed an executive order directing certain AI companies to voluntarily submit new models for government testing and evaluation before public release.

Parallels with Anthropic’s Cautious Approach

This development mirrors the strategy of competitor Anthropic, which earlier this year sparked debate by announcing its frontier cyber model, Claude Mythos, would only be released to a small group of partners through a program called Project Glasswing. Anthropic argued the model was too powerful and could cause significant harm if misused. Critics and observers have since debated whether Anthropic’s stance is a genuine safety measure or a marketing tactic.

The core concern with frontier AI models like GPT 5.6 and Claude Mythos is their ability to identify and exploit software vulnerabilities at speeds far beyond human capability. LLMs have already proven effective at writing malware, and some can autonomously execute ransomware attacks. For organizations running complex software infrastructure, this presents a clear and significant security challenge.

What This Means for the AI Industry

The White House’s involvement signals a potential new normal for AI governance in the United States. While the administration has not yet imposed mandatory regulations, its willingness to request limited releases suggests a more interventionist posture than previously expected. For companies like OpenAI and Anthropic, this creates a delicate balance between innovation, competitive pressure, and government relations.

Because these frontier models remain closed to the public, it remains difficult to independently verify the extent of the risks they pose. The lack of transparency makes it challenging for the broader tech community to assess whether the government’s caution is proportionate or overly restrictive.

Conclusion

The Trump administration’s request for OpenAI to slow the release of GPT 5.6 marks a significant moment in the evolving relationship between the U.S. government and leading AI developers. As both companies and regulators navigate uncharted territory, the outcome of this limited release could set a precedent for how future powerful AI models are brought to market. The coming weeks will reveal whether this cautious approach becomes a permanent fixture of the AI landscape.

FAQs

Q1: Why is the White House asking OpenAI to slow the release of GPT 5.6?
The administration has requested a limited release due to safety concerns about the model’s potential to be misused, particularly in cybersecurity contexts. The government wants to evaluate the model before broader public access.

Q2: How is this different from previous OpenAI releases?
Historically, OpenAI has released models widely and quickly. For GPT 5.6, the company plans to approve access on a customer-by-customer basis during a preview period, with government agencies involved in the process.

Q3: What are the main risks associated with frontier AI models like GPT 5.6?
These models can identify and exploit software vulnerabilities at high speed, write malware, and autonomously execute ransomware attacks. This poses significant risks to organizations with complex software infrastructure.