📰 TL;DR

Anthropic beat OpenAI for the first time in enterprise spending share at the end of May, data from Ramp shows covering over 70,000 business clients. Simultaneously, Anthropic closed a $65 billion funding round at a $965 billion valuation, also surpassing OpenAI, and quietly filed an IPO - reportedly driven by the company’s first profitable quarter.

However, in early June, the Trump administration sent a letter to Anthropic, invoking a obscure export control rule, demanding that non-US persons (including Anthropic’s own employees) be denied access to its latest flagship models: the restricted release Mythos 5, and Fable 5, which was publicly released just three days prior. This effectively forced Anthropic to pull both latest models offline. Speculation suggests the move was triggered after hackers easily bypassed Fable 5’s safety guardrails and accessed the full capabilities of Mythos. Mythos is extremely capable at detecting software security vulnerabilities - Anthropic itself had publicly flagged the model as “dangerous” and restricted public use.

This isn’t the first clash. In March, the Trump administration blacklisted Anthropic over “supply chain risk,” after Anthropic refused to let the government use its models for mass surveillance and fully autonomous weapons operation. Ramp’s chief economist Ara Kharazian noted that historical data shows controversy actually boosts commercial adoption - Anthropic’s best enterprise adoption month was exactly when the Defense Department flagged it as a supply chain risk. In May, Anthropic’s AI subscription share rose from 38.5% to 41%, while OpenAI stayed flat at 39.5%. Kharazian put it plainly: “Being labeled a dangerous model comes with its own光环.”

💬 JudyAI Lab Take

Anthropic beat OpenAI in enterprise spending share at the end of May, closed a $65B funding round, and quietly filed an IPO - all three fronts (commercial, capital, regulatory) converging at once. That’s unusual.

The counterintuitive part: the month the Trump admin blacklisted Anthropic over “supply chain risk” was exactly when it saw its strongest enterprise adoption. As Ramp’s chief economist said, “Being labeled a dangerous model comes with its own光环.” The B2B market logic works like this - official warnings actually send the signal “this model really has teeth.” Another red flag: Fable 5 got bypassed by hackers, who then accessed the restricted flagship model Mythos 5. Controlling model capability boundaries is an engineering AND policy challenge - not something a few terms of use can solve.

Next time you see an AI company get flagged over security or regulatory concerns, watch the enterprise adoption numbers the following month - that’s usually the real market verdict.

📅 Source Info

🔗 Further Reading