|
Today, AI Twitter (yes I know it’s “X” now) is abuzz with the news that OpenAI is moving ahead with a staggered release of its latest flagship AI model, GPT-5.6, with the White House reportedly granting approval for customer access to the model on a company-by-company basis.
This is remarkable, and if you’ve been following Control Plane, you know why: We’ve been tracking the trend toward fulfilment of Leopold Aschenbrenne’s prophecy that the frontier AI labs will inevitably be nationalized when the national security implications of their work become more obvious. This seems like another step in that direction, coming shortly after the US government effectively made Anthropic pull its own frontier model, Fable, from public access due to cyber-safety concerns.
Today, just for fun, I’d like to parse the wording of OpenAI’s official announcement, because I found some parts of it interesting. It’s quite diplomatic, and yet fairly open about its underlying message, in my opinion.
The first paragraph is just a standard introduction of the product:
We're beginning a limited preview of the GPT‑5.6 series: Sol, our flagship model; Terra, a balanced model for everyday work; and Luna, a fast and affordable model. Terra has competitive performance to GPT‑5.5 while being 2x cheaper and Luna brings strong capability at our lowest cost.
Pretty normal, especially the confusing change to their naming conventions. That’s right out of the standard OpenAI playbook. But in the second paragraph, they immediately pivot to safety concerns:
GPT‑5.6 Sol launches with our most robust safety stack to date. We strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse, and spent multiple weeks finding weaknesses, pressure-testing our system, and hardening it against real-world attacks.
This is clearly top of mind for folks at OpenAI, and they also want to clearly signal this to everyone else – that they’re taking it super-seriously.
In paragraph three, things get more interesting:
We believe in broad access, and we plan to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks. As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch. At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly. During this preview, we will continue testing and coordinating closely with partners as we work toward broader availability. We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.
There are a few things here that I find interesting.
First, the emphasis on “ongoing engagement with the U.S. government”. It’s a way of saying, ‘Look, we’ve been trying to figure out the safety stuff with the government for a while, and we’re still doing that now.’ It could be read as a slight dig against their arch-rival, Anthropic, which is perceived to have been more combative with the Trump administration, or at least not very engaged. This is, in my view, smart signaling about political savvy.
That and the following sentence are also good marketing. OpenAI is saying, ‘We’ve been engaging with the government in good faith, and so we showed them our latest model. And they were so in awe of its capabilities that they asked us not to release it widely.’
They’re letting the White House make the pitch for them. Of course, they kind of had to. Imagine if instead they said, ‘We previewed it for the government, and they told us we can just go ahead and release it. This model isn’t expected to be a threat to anyone.’
The last few sentences of that section are my favourite, though, because they get sort of openly calculating. OpenAI says they don’t like this approach to regulating model access, but they’re doing it anyway as “a short-term step” because it’s the shortest route to broad availability in this case. It’s sort of like, ‘We think this is stupid but we’re playing along because it’s just the easiest way to move forward right now.’
Of course, they signal that they believe there are better frameworks for this sort of thing, and hopefully continued engagement with the White House will lead to one of those frameworks being adopted.
Notably, a former AI advisor to the White House who recently joined OpenAI, has done some pretty empathetic rationalizing about the government’s behaviour back in April, when rumours first emerged that the release of Anthropic’s Mythos would be blocked. Dean W. Ball previously served as Senior Policy Advisor for Artificial Intelligence and Emerging Technology at the White House Office of Science and Technology Policy (OSTP), and was hired as OpenAI’s Head of Strategic Futures this month. Here’s part of what he had to say on X, the everything app:
1. Assuming the story is true, I suspect the White House is making the right call. But this is the opposite of a tenable strategy, like trying to erect a dam against a tsunami. There is no way to stop the diffusion of capabilities like Mythos within the next 6-18 months. 2. We should be clear that the government restricting the release of AI models is a type of licensing regime. It is an informal, highly improvised licensing regime, but a licensing regime nonetheless. This isn’t going to be the last such model we see of this capability tier, and cyber vulnerability discovery is very far from the only type of dangerous capability. If the government is going to insist on restricting frontier capabilities for the foreseeable future, it will need to formalize the rules for those restrictions—how long must you delay, what objective factors generate a “green light,” etc. I know this will feel even more regulatory to some, but the alternative is an unpredictable, inconsistent, improvisatory licensing system, and this is both bad for business and the rule of law.
In other words – if I may do a little more parsing – he’s saying, ‘Look, the White House doesn’t know what to do because there’s no playbook for this. They are just trying to do something. It will be a lot less chaotic when we have a playbook.’
|
