If you wanted to write a case study on accidental brand-building, Anthropic's May and June would be a pretty compelling first draft.

Let's run the tape. Anthropic closed May by overtaking OpenAI in business spending market share for the first time—data courtesy of Ramp, which tracks AI expenditure across more than 70,000 companies on its platform. It raised $65 billion at a valuation flirting with the trillion-dollar stratosphere, edging out OpenAI there too. Then it quietly filed IPO paperwork, reportedly off the back of its first profitable quarter in company history.

Then the Trump administration decided to lob a grenade into the party.

The Ban That Launched a Thousand Signups

On a Friday—because of course it was a Friday—the White House sent Anthropic a letter demanding it cut off non-American users, including its own employees, from accessing its most capable models: the limited-release Mythos 5 and the freshly-launched Fable 5, the public-facing version of Mythos that had been available for all of three days. The administration cited an obscure export control directive, though the actual reasoning remains murky.

The leading theory floating around? That security researchers had managed to jailbreak Fable 5's guardrails—the safeguards meant to wall off Mythos' full capabilities from the general public. And Mythos' full capabilities are genuinely alarming: the model is apparently so competent at identifying software vulnerabilities that Anthropic had already been marketing it as potentially dangerous and restricting its own distribution. That's not a PR move. That's a company that actually ran its safety evals and didn't like what it found.

So Anthropic pulled Mythos and Fable 5 off the market entirely. Collateral damage from the export control order.

The Supply-Chain Risk That Became a Sales Driver

Here's where it gets interesting. This isn't the first time Anthropic has found itself in the government's crosshairs. Back in March, the Trump administration officially designated Anthropic a supply-chain risk—a designation that followed Anthropic's refusal to let its models be used for mass surveillance of American citizens or fully autonomous weapons systems. Bold stance. Commercially inconvenient, you might think.

Except Ramp's data tells a different story. Ara Kharazian, Ramp's lead economist who compiled the business-spending figures, told reporters: "Anthropic's best month on record, as far as business adoption, was the month that the Department of Defense labeled them a supply-chain risk. There's a lot of aura that comes with your model specifically being named too dangerous to use."

Let that sink in. Getting called a national security concern was, functionally, a demand generation event.

Kharazian's prediction for this latest bout of government drama? "If anything, it'll probably boost them."

What the Numbers Actually Show

By May, Anthropic commanded 41% of AI subscription spending among Ramp's business customers—up 2.5 percentage points in a single month. OpenAI sat at 39.5%, essentially flat. That's a meaningful shift, even if OpenAI still dominates consumer-side usage by a wide margin (Sensor Tower's data makes that clear).

But subscriptions are almost a rounding error compared to API spend. The real money is in token consumption—businesses hammering the API for coding assistance, document processing, and whatever other workflows they've wired Claude into. Anthropic's Claude Code has built a genuine reputation as a serious coding tool, not just a demo that falls apart under production load.

When Ramp can actually see which models are being called—about a third of transactions have that granularity—businesses are overwhelmingly reaching for Claude Opus, especially the more recent versions. Opus 4.8 dropped in late May, and it's still fully available. Mythos was barely on the market before it got yanked—limited release since April, Fable 5 lived for fewer days than most leftovers in your fridge.

The IPO Wrinkle Nobody Can Ignore

Here's the part where I'll temper the triumphant narrative slightly. Public market investors are a different breed from enterprise buyers. Enterprise buyers look at a "too dangerous to use" designation and think powerful tool, good moat. Institutional investors preparing to price an IPO look at ongoing regulatory conflict with the federal government and start asking uncomfortable questions about revenue predictability, legal exposure, and whether the CEO is going to be called before a Senate committee next quarter.

Ramp's data doesn't have the resolution to tell us exactly how much revenue Anthropic loses from pulling its flagship models. That's a real question mark heading into any public offering.

What the data does tell us: the models that are still on the market are more popular with businesses than they've ever been. The core business is growing. And somehow, every time the government tries to make Anthropic look dangerous, enterprises respond by opening their procurement portals.

There's probably a dissertation in there somewhere about how AI safety credibility became a B2B growth lever. For now, Anthropic seems content to let the controversy do its marketing for free.