Selective Safety: Why the White House Targets Anthropic but Spares OpenAI

Two weeks ago, I argued that Anthropic had walked into a safety trap of its own making: by hobbling its Mythos model for show, it invited the US government to take performative safety seriously. This week, the punchline arrives with a twist. OpenAI released an update to its GPT-5.5-Cyber model with capabilities that, by its own admission, rival Anthropic's Mythos. The White House's response? A thundering silence. No export controls, no national security directives, not even a terse press release. The administration that rushed to restrict Anthropic's model has apparently found nothing to say about OpenAI's equivalent.

Let us parse this discrepancy. Anthropic, the company that literally wrote the book on safety — or at least a very long paper about it — gets its foreign access suspended. OpenAI, the company that released ChatGPT into the wild without so much as a content warning, gets a free pass. The difference cannot be technological; OpenAI's model is described as a rival in capability. The difference is strategic. Anthropic made itself a target by screaming about safety from the rooftops. It hobbled its own model to prove a point, then donated $20 million to a Super PAC to shape AI policy. Meanwhile, OpenAI keeps its head down, releases powerful models, and lets its investors lobby quietly. The message is clear: if you talk about safety, you become a target. If you say nothing, you're fine.

This is not safety regulation. It is political convenience. The Trump administration, ever eager to flex its national security muscles, needed a scalp. Anthropic, with its public hand-wringing and theatrical hobbling, offered itself up on a platter. OpenAI, by contrast, has cultivated a relationship with the administration that apparently insulates it from scrutiny. The $20 million Super PAC donation from Anthropic now looks less like influence-buying and more like painting a bullseye on your own back. You can't pay your way out of a narrative you started.

And what of the collateral damage? A startup named Legion has sued the US government over the export control directive that blocks its access to Anthropic's Fable 5. The lawsuit argues that the restriction is illegal and that the model is critical to its business. This is the inevitable consequence of safety theatre: real companies, real livelihoods, caught in the crossfire of a regime that treats AI restrictions as political weapons rather than evidence-based protections. The irony is rich. Anthropic's gambit to prove its safety bona fides has, in practice, made the technology less accessible for legitimate commercial use, while OpenAI — which has said far less about safety — continues to operate unencumbered.

The lesson for the industry is depressingly obvious. The game is not about making AI safer; it's about managing appearances. If you draw attention to risks, you become the risk. If you quietly release powerful models, you're a patriot. The real danger isn't that AI is unregulated; it's that regulation is wielded as a cudgel against the naive, while the clever ones get a free pass. Anthropic's tragedy is that it believed its own rhetoric. Everyone else should take note.