# OpenAI to release delayed models Thursday amidst a sea of regulatory confusion > Source: > Published: 2026-07-08 18:59:27+00:00 As enterprises struggle to manage their AI strategies, the US AI regulatory environment is sending a wide range of contradictory signals. OpenAI’s Wednesday announcement that it will now release GPT-5.6 Sol, along with Terra and Luna, on Thursday highlights the confusion. Initially, the US government said that it was asking OpenAI to [limit access to its top models](https://www.infoworld.com/article/4190089/us-tells-openai-to-restrict-access-to-its-most-powerful-ai-model-2.html), including the three releasing Thursday, to a short list of companies. OpenAI seemingly agreed and held back their general availability. But on Wednesday, OpenAI reversed its position, with [a statement on X](https://x.com/OpenAI/status/2074704958419792299?s=20) saying simply: “GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday. We’re expanding preview access globally now.” No details were released about the extent of the expansion. Then the White House issued a statement, a copy of which it emailed to *InfoWorld*, saying that the US government “did not give OpenAI a ‘green light,’ approval or clearance to release its models. No such permission is required or granted. The Administration does not provide approvals for private companies to release AI models – decisions on timing and scope of releases rest entirely with the companies.” The statement then quoted from the [June 2 White House executive order](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/) that said, “nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.” It also said, “any testing or meetings with government experts is voluntary. Participation is not required to release a model.” Yet last month, the US Commerce Department weighed in [on how Anthropic’s models can be distributed](https://www.cio.com/article/4186429/anthropic-fable-dispute-suggests-export-no-longer-means-what-it-used-to.html). [Lewis Carhart](https://www.linkedin.com/in/lewiscarhart/), CEO of software development firm Comp AI, said the statement is frustrating for IT executives on multiple fronts. “Think about what that [White House statement] means. OpenAI stationed engineers in Washington for weeks, submitted to government testing, staggered its launch at the government’s request. And the official position is that none of that was required,” Carhart said. “We now have a de facto licensing regime that legally doesn’t exist. There’s no statute, no appeal process, no published criteria. Just [the US Department of] Commerce deciding model by model what ships and when.” That’s the worst of both worlds, he noted: “All the friction of regulation with none of the predictability. Compliance people have a name for this – it’s an audit with no framework. And the precedent is now locked in for both frontier labs: if you build at the frontier, your launch calendar runs through Washington whether the law says so or not.” Carhart argued that this regulatory reality should be of extreme concern to enterprise IT executives, given it indicates that model availability is now “a regulatory variable” not driven by the vendor roadmap. “Anthropic’s most advanced models disappeared from the market for three weeks in June. It was not because of an outage, not because of a pricing change. It was because of an export control directive,” he pointed out. “If your AI architecture assumes the model you deployed today is available tomorrow, that assumption is now demonstrably false. Multi-model resilience just went from nice to have to a board-level risk item.” It also offers an opportunity, given that a model that cleared government security testing is a model that auditors and boards sign off on faster. “Government review is quietly becoming a procurement asset,” he observed. “The CIOs who win here are the ones who treat ‘regulatory posture of the model itself’ as a line item in vendor risk assessments – most risk teams are still only looking at the provider’s SOC 2 attestation.” [Jason Andersen](https://moorinsightsstrategy.com/team/jason-andersen/), principal analyst at Moor Insights & Strategy, agreed with Carhart and described the overall back-and-forth as “a bit of pageantry. OpenAI needs its model to look as powerful, potentially dangerous, as Anthropic’s so it can be a contender to be at the absolute frontier. It also helps to burnish OpenAI’s PR efforts to look more responsible than it has in the past.” Furthermore, he added, “tech CEOs are acutely aware that flattery towards this administration could keep regulators or the threat of regulation at bay.” [Brian Jackson](https://www.infotech.com/profiles/brian-jackson), a principal research director at Info-Tech Research Group, echoed the political concerns. “What’s still not clear is the actual criteria being used to deem the models safe for release. The government has said that it’s concerned about cybersecurity risk as well as the risk of AI being used to develop biological weapons,” he said. “But so long as the actual release criteria lack transparency, there will be some perception that the evaluation could be politically motivated.” Jackson said that one, presumably unintended, result of the US government’s efforts to control AI rollouts is that it is making companies look far more seriously at using non-US vendors for AI strategies. “Organizations are looking for alternatives to the private US-based cloud-delivered frontier models. That’s so they can maintain control over their AI supply chain,” he said. “Chinese open source models are one option that’s available, but there are other options too, from Canada and Europe. Companies can either set up AI access through other APIs not connected to US-based AI providers, or download open-source models to run locally.” He noted that the added US regulatory risk means that some organizations will avoid becoming entrenched within OpenAI’s and Anthropic’s interfaces, where there’s no option to swap out the LLMs for an alternative. [Rock Lambros](https://zenity.io/authors/rock-lambros), director of AI standards and governance at AI agent vendor Zenity, shared the frustration that little to no actionable compliance data is being released. “Nobody outside a closed room can tell you what standard [the model] passed because it was never written down. For two weeks, [US government officials] kept a model out of defenders’ hands that’s better at guarding your network than breaking into anyone else’s,” Lambros said. “Call that a security review if it helps you sleep better. But it reads to me like a bouncer working a velvet rope nobody hired him to run, waving people through today because he’s in a better mood than he was a couple of weeks ago.” This unpredictability is likely to have impacts on AI strategy far beyond traditional compliance concerns, Lambros said. “We’ve built way too much operational reliance on these models to hang it on a review with no rulebook,” he said, pointing out that hospitals, pipelines, banks and water utilities are relying on frontier AI whose availability “can swing from ‘on’ to ‘off’ to ‘on’ with no notice, no appeal, and no published standard behind any of it.” “You can’t run critical infrastructure on a tool that runs fine Friday and is offline by Monday because an approval process nobody can see reached a verdict nobody can predict,” Lambros said. “That is a supply chain risk with a government hand on the switch, and almost nobody has priced it into a continuity plan.” To protect themselves, companies need to adjust their expectations. “Treat model availability like any single point of failure you don’t own by standing up a fallback you’ve tested, getting a continuity clause in your contract, and drilling for the blackout, because ‘the government backed off this time’ is not a plan,” he advised. *This article originally appeared on InfoWorld.*