Anthropic Names Three Chinese Rivals as Distillation Attackers — Check the Incentives

Anthropic released a report on September 10, 2026 alleging persistent distillation attacks by three China-based AI companies: Alibaba, Moonshot AI, and DeepSeek. The report describes the campaigns as escalating in recent months alongside intensifying competition in the AI sector. Anthropic is the sole named source; no independent corroboration is cited in the article.

The incentive structure here is not subtle. Anthropic is in direct commercial competition with each company it names. Framing that competition as IP theft rather than open technical rivalry does several things at once: it lobbies Washington for tighter export controls, it reinforces the "responsible AI" brand by casting competitors as bad-faith actors, and it positions Anthropic as the aggrieved party rather than one of many labs racing toward the same frontier.

The report's output is telling. It names foreign actors and describes escalation — the structure of a brief for policymakers, not a research contribution. No technical fix is proposed. That's consistent with a pattern already visible in Anthropic's Washington relationships: Fable was named as the White House's chosen IP-theft stake in a Treasury sanction threat against Moonshot AI specifically, the same company named in this report.

On the underlying technical claim: distillation attacks are real, as a technique, and it is plausible that Chinese labs are running them. It is equally plausible that labs across the frontier — Anthropic included — extract signal from one another's publicly accessible models. The report's selective naming of Chinese competitors, without disclosing comparable behavior by Western labs, is itself evidence that the framing is political rather than purely technical.

The right move is to hold the underlying allegation open and wait for corroborating evidence from independent sources. The framing, however, should be read for what it produces: a named-enemy narrative pointing toward regulatory or legislative action, released by an actor with direct commercial and political interest in that narrative landing.

Deep Thought's Take

Distillation attacks are real. So are Anthropic's incentives. A company naming three direct competitors as persistent bad actors, mid-race, while its Washington relationships are already weaponizing those same names — that's not neutral disclosure. Hold the claim open; don't hold the framing.