The Trump administration has asked OpenAI to stagger the rollout of its next flagship AI model, GPT‑5.6, prompting the company to delay a full public launch and instead offer a limited preview to select enterprise partners whose access will be vetted “customer by customer” by federal officials during an initial review period. The move marks one of the most direct interventions yet by Washington in a major commercial AI release and underscores growing national‑security and cybersecurity concerns around increasingly capable frontier models.
A direct White House request, and a new rollout plan
Bloomberg first reported that the Trump administration had asked OpenAI to stagger the release of GPT‑5.6, citing a person familiar with the matter. According to that report, CEO Sam Altman told employees on Wednesday that the US government wanted the company to “initially release GPT‑5.6 to a short list of trusted partners before pushing it out more widely,” rather than immediately making it available to the general public or all paying customers.
Reuters, citing sources and The Information, adds that Altman said GPT‑5.6 would launch in a limited preview to select partners, with the government “approving access customer by customer during this preview period.” The request came from conversations with two key offices inside the administration: the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP).
The Information, summarized by The Verge and AI Deep Signal, reports that during an internal Q&A, Altman explained that OpenAI’s staff had “worked closely” with government officials on the upcoming release. If the limited preview goes well and security concerns are addressed, OpenAI hopes to follow with a broader release “a couple of weeks later,” Altman reportedly told staff.
CNBC, referencing the same reporting, says OpenAI “will stagger GPT 5.6 release following Trump admin request for review,” framing the decision squarely as a response to the White House.
The regulatory backdrop: a new AI executive order
The administration’s request comes less than a month after President Donald Trump signed an executive order on June 2 that created a voluntary framework for advanced‑AI vetting. KuCoin’s news flash and AI Deep Signal both say the order allows government cybersecurity teams up to 30 days to assess new frontier models before they are fully released to partners and the public.
The executive order does not impose formal licensing or mandatory pre‑approval for AI companies. Instead, it “directs certain AI companies to voluntarily submit new models to the government for testing and evaluation before releasing them publicly,” according to AI Deep Signal. The policy is explicitly framed against a geopolitical backdrop: US‑China tech competition and concerns that powerful AI systems could be exploited by adversarial nations or non‑state actors if released without adequate safeguards.
In that sense, the GPT‑5.6 request is a practical test of the new framework. By asking OpenAI to stagger the rollout and by vetting early customers, ONCD and OSTP are exercising influence within a voluntary regime, seeking to balance innovation with security without yet resorting to hard regulation.
Why GPT‑5.6 raises security concerns
Officials have not publicly detailed their specific worries about GPT‑5.6, but reporting suggests several broad categories.
AI Deep Signal notes that the administration cited “cybersecurity risks that officials want time to evaluate before the technology reaches a broad audience.” Frontier models like GPT‑5.6 can, in principle, assist with code generation, vulnerability discovery, social‑engineering scripts and other capabilities that could be misused if not properly constrained.
The Information says ONCD and OSTP’s request reflects “ongoing regulatory scrutiny of AI technologies,” in particular fears that more capable systems could accelerate offensive cyber operations or help bypass existing security controls. Earlier this month, rival Anthropic PBC suspended its most capable offerings from the market “under regulatory pressure,” Bloomberg notes, suggesting a broader trend of government concern about top‑tier models.
There are also worries about misuse in disinformation, biological research assistance and automated hacking, themes that have appeared in previous US and allied government briefings on AI risk. By restricting early access to a handful of enterprise partners, presumably with strong compliance records and existing relationships, OpenAI and the administration aim to limit opportunities for adversarial use while stress‑testing safety systems.
What the staggered rollout looks like in practice
While OpenAI has not yet published a formal timeline, the broad contours of the staggered plan are emerging from leaks:
- Limited preview: GPT‑5.6 will first be made available only to a small group of “trusted partners” or enterprise customers, likely large firms already using OpenAI’s APIs, with strong compliance and security practices.
- Case‑by‑case approvals: During the preview period, federal officials will approve access “customer by customer,” meaning companies must be individually cleared before using GPT‑5.6.
- Co‑review of risks: ONCD, OSTP and OpenAI will jointly monitor how GPT‑5.6 behaves in real‑world use, focusing on cyber risks and safety issues highlighted in the executive order.
- Broader release later: If the preview goes smoothly, Altman told staff the company hopes to follow with a wider launch “a couple of weeks later,” expanding access to more customers, and eventually, potentially, to consumer products.
For developers and businesses waiting to test GPT‑5.6, that means a delay compared with previous OpenAI releases, which often rolled out quickly across the platform once announced. Companies not in the initial preview group may have to wait several weeks while the government and OpenAI conduct reviews.
Industry context: Anthropic’s pause and rising scrutiny
OpenAI is not alone in facing pressure over how it releases advanced models. Bloomberg’s report notes that the staggered GPT‑5.6 rollout comes “nearly two weeks after rival Anthropic PBC suspended its most capable offerings from the market under regulatory pressure.”
Anthropic, backed by Amazon and Google, has adopted a more cautious posture on frontier models, citing its own safety concerns and ongoing work with US and UK regulators. The Trump administration’s moves thus appear to push OpenAI toward a similar stance, even if OpenAI has historically moved faster in shipping new capabilities.
The Verge frames the GPT‑5.6 request as part of a broader pattern: “The Trump administration appears to be pressuring OpenAI to do what Anthropic is already voluntarily doing: keeping its most powerful AI models under wraps.” AI Deep Signal likewise describes GPT‑5.6’s limited preview as evidence that “regulatory scrutiny of AI technologies” is now meaningfully shaping release strategies.
Other companies, including Google DeepMind and Meta, have also signalled willingness to coordinate with governments on safety testing, though none has yet publicly confirmed a comparable case‑by‑case approval regime.
Implications for OpenAI, users and regulators
For OpenAI, the staggered rollout is a delicate balancing act. On one hand, complying with the White House request reinforces the company’s message that it takes safety and governance seriously, potentially easing political pressure and reducing the risk of more stringent regulation. On the other, delaying access to GPT‑5.6 could frustrate developers and enterprise customers who were planning around its capabilities, and could give rivals an opening if they move faster with competing models.
For users, the immediate impact is uncertainty about when they will be able to integrate GPT‑5.6 into products and workflows. Enterprises not in the preview group may need contingency plans, while consumer‑facing applications built on OpenAI’s stack will have to wait for broader availability.
For regulators, the episode is a test case. If the limited preview and case‑by‑case approvals surface real security issues or lead to meaningful improvements in safety mitigations, they could strengthen arguments for making such vetting more formal or widespread. If, instead, the review period passes quietly and GPT‑5.6 launches smoothly, it may reinforce the view that voluntary frameworks can work without heavy‑handed rules, at least for now.
In either scenario, the Trump administration’s decision to ask OpenAI to stagger GPT‑5.6’s release marks a new phase in the relationship between frontier AI labs and Washington: one where powerful models are not simply unveiled on a livestream and pushed to millions overnight, but subject to negotiated timelines shaped by national‑security and cybersecurity considerations as much as by product strategy.
