Calling for options around a possible ban/pause in the future is very different from calling for a ban/pause now. But it's a clear and reasonable goal. That people confuse these two different things doesn't make either of them less clear on its own.
If you expect likely extinction/disempowerment on the current trajectory, a mere option for a pause can sound like hedging, but it's still directionally useful. If you expect a good outcome of further AI development when taking feasible precautions, then an option for a future pause is one of the precautions that you would like to take, but a ban/pause now would instead be an opportunity cost catastrophe.
So there should be broader agreement in developing options for a ban/pause, while demanding that options for a pause must come with declarations of support for a ban/pause as soon as possible is in conflict with those who don't think a pause as soon as possible would be a good thing. Perhaps such conflict is appropriate, if you think that it's already somewhat late for a pause, that waiting for concrete and actionable signs of things going very wrong would find that the pause buttons no longer work.
This is how Anthropic and other AI companies can signal safety to one room, acceleration to the next, and out-accelerating every rival to its investors, all at once. ... A good lobbyist puts on a concerned face, shows deep worry about those impacts, and promises to keep doing exactly what his company is doing already, but “responsibly”.
There is disagreement on facts, leading to disagreement on policy. Being somewhat evasive and non-confrontational when expressing this disagreement (as opposed to stating it plainly, as a conflict on policy) is not the same as manipulation with misleading messaging.
A big argument in this post is that Anthropic's communications are overly vague, yet it backs this claim up with only quoting two sentences from the RSI blog post.
In the original RSI post, Anthropic actually goes into considerable detail into their position and why they choose their stance given the competitive dynamics that are involved. It is true that they didn't make hard commitments, but that is fully explainable by the uncertainty and competitive dynamics involved.
The other point here is about the apparent contradictions ("muddying the waters") this post tries to convey about Anthropic employees' stated opinions in public appearances. This section reads like bogus to me, as it is perfectly reasonable that one is excited about AI's benefits and simultaneously wants to reduce the drawbacks, and expresses both to different audiences to find common ground.
They will use this to continue placating all sides, from the accelerationists to the safety-concerned.
However, little will change in terms of actions. The companies are continuing to pursue superintelligence at full speed.
There is a logical contradiction here. Why would AI companies need to "placate all sides" including the accelerationists (per the first sentence), if they are already going full accelerationist anyway (per the second sentence)?
Last week, the AI company Anthropic released a blog post titled “When AI builds itself”. This led to a media frenzy, with news outlets around the world publishing headlines that the company was urging a global pause on AI development, or calling for AI non-proliferation.
However, the post does not call for a pause. The post warns that the self-improving AIs that Anthropic is developing could “increase the risks of humans losing control over AI systems” and says “it would be good for the world to have the option to slow or temporarily pause frontier AI development”.
Anthropic’s blog post’s language is deliberately vague, underscoring that companies will not lead a slowdown.
Just yesterday, OpenAI followed suit with a post co-authored by their CEO, Sam Altman, and Jakub Pachoki, a company executive in charge of recursive self-improvement. This post was met with similar, albeit smaller, reception as the Anthropic post, with some stating that Altman is calling for a slowdown in AI development.
Once again, the post in question does not call for a slowdown in AI development. The post contains vague language about the need of international coordination, “One goal of such an organization should be to make it possible for the world to take coordinated action, including slowing frontier development when needed”, echoing the hedging language of Anthropic.
This is not an isolated case. It’s a deliberate PR approach that Anthropic and OpenAI have used over and over to curry favor with multiple opposing audiences, while making no concrete commitments.
The approach consists of the following:
Muddying the waters and playing all sides
At the heart of the Anthropic post is their work to achieve AI recursive self-improvement (RSI), where AI systems improve other AIs without human involvement, the most direct path to superintelligence.
Anthropic’s own chief scientist, Jared Kaplan, has called allowing recursive self-improvement the “ultimate risk” and warned it could be the moment humans lose control of AI.
Co-founder Jack Clark gestures at some of the recursive self-improvement risks in the post too, yet watch how Clark handled it on CNN once the post was out. Asked by Anderson Cooper about RSI’s upside, he described it as creative co-scientists advancing medicine, biology, and robotics.
Asked about the downside, Clark sidestepped that very loss-of-control scenario and moved to talking about how we could verify and trust these systems, comparing it to dropping “hundreds or thousands of new colleagues” into the newsroom.
And when Anderson Cooper asks whether Anthropic wants to see the industry as a whole slow or pause AI development, Clark’s answer opened bluntly: “Our view is we’ve built amazingly powerful technology. We’re going to keep building it.” Two years ago, the company’s CEO Dario Amodei called pausing AI development the “most extreme” version of an “extreme position”, and “the opposite” of Anthropic’s policy.
Anthropic’s approach is OpenAI’s approach
This approach is not unique to Anthropic. Anthropic is replicating OpenAI’s techniques of misdirection and deflection. This has been surprising for many people, who see Anthropic as a special case among AI companies. However, Anthropic deploying these techniques is to be expected, given that all of Anthropic’s cofounders were previously at OpenAI.
Dario Amodei was involved with OpenAI since its foundational days, one of the few people at the 2015 dinner between Sam Altman and Elon Musk that led to the company’s creation. Amodei then led AI research at OpenAI for five years before spinning out his own company in 2021.
Jack Clark joined OpenAI in its early days and built up its media relations, PR, and lobbying functions, training a generation of lobbyists at the company. He then followed Amodei and others in co-founding Anthropic in 2021, where he similarly leads their lobbying and PR efforts.
OpenAI has applied the same technique for years. In 2023, Altman testified to Congress about the importance of regulating AI, giving the safety-concerned a statement to point to. By 2025, signaling to a different audience, OpenAI was warning that regulation would blunt America’s competitive edge over China, and rejecting calls for it in a further hearing.
The same pattern of switching back and forth between positions runs through OpenAI’s stance towards state regulation of AI. Across 2025 OpenAI pushed to preempt state AI rules, then, once Mythos turned Washington toward scrutiny, the company endorsed Illinois state legislation in spring 2026, including a bill that would largely shield it from liability. None of these positions bound the company to change its course. Each gave a different audience a position it could welcome, while leaving OpenAI free to continue business as usual.
OpenAI did all of this while continuing to develop superintelligence, a technology Altman acknowledges poses an extinction risk on par with nuclear war and that he called “probably the greatest threat to the continued existence of humanity”.
OpenAI’s latest essay, just published yesterday, uses the same approach. The post includes a vague mention of international coordination, all the while being very clear on what OpenAI’s goals are, including “build an automated AI researcher” to have AIs automate AI R&D: recursive self-improvement.
This is how Anthropic and other AI companies can signal safety to one room, acceleration to the next, and out-accelerating every rival to its investors, all at once.
These techniques are not new to AI companies either: they have a storied history of being used by major corporations developing dangerous products. The oil industry pioneered plausible deniability to keep lead in gasoline as early as the 1920s, and Big Tobacco refined it from the 1950s onwards, spending decades obscuring their products’ damaging impacts even when the harm was well known internally. A bad lobbyist categorically denies his company’s products harm people. A good lobbyist puts on a concerned face, shows deep worry about those impacts, and promises to keep doing exactly what his company is doing already, but “responsibly”. The former gets regulated; the latter is hailed as a thought leader.
Don’t get got
So what is the takeaway from these new talking points from AI companies? Not that these companies keep changing their minds. The pattern is deliberate, and lobbyists are going to lobby. The lesson is to look at the facts, and look at the source material.
Watch out for communications that are very vague, as Anthropic’s and OpenAI’s hedging statements about the option to pause or slow down AI development are. Do not take deliberately vague statements as wins.
Pay attention to actions and unequivocal statements. While Anthropic and OpenAI’s language on pauses and slowdowns is hedgy and vague, Anthropic is very clear about other matters. For instance, their commitment to do recursive self-improvement. They have consistently been pitching recursive self-improvement to their investors as their path to achieve a decisive strategic advantage, have dedicated vast amounts of talent and resources to pursue it, are hiring world famous AI scientists to automate AI development, and are unequivocally stating their pursuit of it in the blog post.
OpenAI is doing the same: while their post is vague on the global coordination part, it very clearly restates their goal of pursuing an “automated AI scientist” to have AIs fully automate AI development.
So are there any implications of them making vague allusions to a pause or slowdown? Yes, obviously. But as explained above, it is not ‘Anthropic and OpenAI will now call for a pause’. In practice, Amodei, and possibly Altman too, might speak less to publicly oppose those calling for a pause or slowdown in AI development, as their talking points have adapted. They will use this to continue placating all sides, from the accelerationists to the safety-concerned.
However, little will change in terms of actions. The companies are continuing to pursue superintelligence at full speed.
The main takeaway of the posts is that both Anthropic and OpenAI are pursuing recursive self-improvement as the most direct path to superintelligence, and making progress on it.
Recursive self-improvement is the fastest pathway for these companies to reach superintelligent AI. Autonomous AI more competent than humans across the board, capable of degrading and subverting the security forces of major nation states. AI that neither they, nor any government on the planet, is able to control.
This threatens the United States and other governments, and threatens human extinction. To prevent this from happening, we need a coalition of countries preventing the development of superintelligence at home and abroad. This means putting an international ban on the development of superintelligent AI in place.
At ControlAI, we’re working on building this international coalition to make the ban happen: if you’re interested in making it happen, get in touch!