So here’s something wild. Anthropic just pulled its most powerful models from the market completely, and somehow their business is booming. That’s not the narrative you’d expect when the government basically tells you to shut down your flagship products, but the data tells a different story.
If you missed it, the Trump administration sent Anthropic a letter demanding they ban non-Americans, including their own employees, from accessing their state-of-the-art models. We’re talking about Mythos 5 (limited release) and Fable 5 (public release, pulled just three days later). The White House invoked some obscure export control directive, but the rumblings in security circles suggest the real issue was that Fable 5’s guardrails were essentially Swiss cheese. Researchers found it way too easy to bypass the restrictions and access the full Mythos capabilities underneath, which is concerning when you’re building AI that can find security flaws in software code. Anthropic themselves marketed Mythos as dangerous and restricted its release. That’s not marketing hyperbole that’s a company essentially saying “we made something we’re not sure we can control.”
The Irony of Being Too Good
This whole situation has a strange irony I’ve been thinking about. Back in March, the Trump administration declared Anthropic a supply-chain risk after the company refused to let the government use their models for mass surveillance of Americans and fully autonomous weapons. Most companies would crater under that kind of designation. Instead, Ramp’s data which tracks business spending across more than 70,000 companies shows Anthropic’s share of AI subscriptions jumped 2.5 percentage points in May to hit 41%. OpenAI sat at 39.5%, essentially flat from the prior month. That’s a genuine lead in the business segment.
Anthropic’s best month on record for business adoption was literally the month the Department of Defense labeled them a supply-chain risk. Ramp’s lead economist Ara Kharazian put it perfectly: there’s a lot of aura that comes with your model specifically being named too dangerous to use. That’s the kind of narrative flip that makes you realize how differently this industry evaluates risk compared to traditional sectors.
What Actually Matters for Developers
Let me break down what’s happening here from a practical standpoint. The vast majority of business spending on artificial intelligence isn’t even on subscriptions. It’s on API calls covering token usage for coding, analysis, and all the day-to-day work developers actually do. Anthropic’s Claude Code has built a genuine reputation as a powerful AI coding tool, and that reputation isn’t built on marketing. It’s built on developers using it and getting results.
When Ramp can see the model details in about a third of transactions, businesses are overwhelmingly spending on various flavors of Claude Opus, particularly the later versions. Opus preceded Mythos and remains openly available. In late May, Anthropic released Opus 4.8, which is still very much in the market. The models that got pulled Mythos and Fable 5 hadn’t even been widely available. Mythos was limited to specific users since April, and Fable 5 lasted all of three days.
So here’s the thing. If you’re a developer building applications today, this news actually has less impact than the headlines suggest. Your existing Claude integrations aren’t going anywhere. The models that businesses actually rely on for production workloads are still there, and usage continues growing. What this does tell us is something more nuanced about where this industry is heading.
The Bigger Picture
We can’t predict exactly how this latest feud will impact Anthropic’s planned IPO. Public market investors tend to be skittish about companies tangled up in government controversies, especially when export controls and national security are involved. But the numbers don’t lie: Anthropic’s available models are more popular with businesses than ever before.
I’ve been thinking about what this means for the broader OpenAI versus Anthropic rivalry too. The consumer market still heavily favors OpenAI according to Sensor Tower data, but in the business segment where companies are actually spending real money on subscriptions and API usage, the gap has closed dramatically. For developers choosing which models to build on, that’s a meaningful signal. Business adoption isn’t just about who has the flashiest models it’s about who institutions trust enough to bet their operations on.
The guardrails debate is maybe the most important takeaway here. The whole reason Fable 5 got pulled is that researchers could bypass them. If you’re building with AI models, you need to understand that “safety features” and “guardrails” aren’t permanent walls they’re evolving challenges. Anthropic marketed Mythos as dangerous and restricted it, and then the guardrails on the public release failed anyway. That’s a sober reminder that as models get more capable, the gap between what’s restricted and what’s accessible gets more consequential.
This whole saga might end up helping Anthropic despite the IPO uncertainty. Sometimes being told you can’t have something makes it more valuable. Sometimes being deemed “too risky” by the government is the best marketing you could ask for. The industry keeps finding new ways to surprise me, and honestly, I think that’s the most interesting part.