The attack window ran from late April through early June, overlapping with Fable 5's pre-launch period. The targets tell the story: agentic reasoning, complex software engineering, long-horizon task completion. The specific capabilities that separate frontier models from everything a few rungs below. The stuff that's hardest to build from scratch and most commercially valuable to shortcut.
A fun detail to sit with: Alibaba's Qwen lab announced a partnership with Fireworks AI on May 1 for "optimized, production-ready deployment" of Qwen's closed-weights models. Smack in the middle of the alleged distillation window.
The phrase "harvesting U.S. AI capabilities" is doing heavy political lifting. It reframes a terms-of-service violation as something closer to industrial espionage, building the case that model reasoning patterns are national assets. And it landed. Export controls followed two weeks later.
For the IPO narrative, being stolen from at industrial scale by a major Chinese tech conglomerate is perversely strong evidence of competitive moat. Thousands of fake accounts spun up to siphon your model is a pretty compelling proof point for investors.
The fallout is already showing up in infrastructure. When Fable 5 came back online after the export control pause, Anthropic had started collecting government IDs and biometrics. One identity verification system solving two problems at once: nationality-based access filtering for compliance and fraudulent account prevention for distillation defense.
Safe to assume most AI platforms are quietly auditing their own account verification stacks right now.