Anthropic Is Fighting Attempts to Copy Its Claude AI Models
Anthropic says new safeguards have cut data harvested in distillation attacks on Claude by about 90% to 99%, banning tens of thousands of accounts and naming Z.ai as an example. The figures are Anthropic's own preliminary results and haven't been independently verified.
Anthropic says it is winning a quiet fight over who gets to copy Claude. The company said Monday that new safeguards have reduced data harvested in so-called distillation attacks by about 90% to 99%, according to Reuters. Distillation means using a leading model's outputs to train a cheaper rival, and it has become a flashpoint in US-China AI competition.
The headline number comes from Anthropic itself, and the company describes the evidence as preliminary. Its claim centers on Claude's protected reasoning, where Anthropic says the number of reasoning tokens extracted by distillers is now "10 to 100 times lower than it was before." A few details show how the defenses work:
- Anthropic restricts the transfer of reasoning logs across conversations, models and accounts, and uses a two-stage classifier to spot and block extraction requests, according to a summary of its post
- The company says the system is "roughly as robust as those that prevent Claude from helping build bioweapons," with a false positive rate under 0.1%
- It has banned tens of thousands of accounts tied to distillation campaigns and cited Chinese startup Z.ai (Zhipu) as an example of a firm that attempted to distill Claude while building its GLM-5.3 model
The caveats are Anthropic's own. It acknowledges the measures have not yet been tested against top-tier threats, that some developers may need to update their integrations, and that the classifier may occasionally block legitimate requests.
Since September 24 it has also resumed charging for requests its safeguards block. The announcement follows a month of escalating accusations, including a US government advisory naming six Chinese labs, and China's Commerce Ministry has called distillation a neutral technique used worldwide.
Anthropic is now asking developers and cloud platforms across the industry to cooperate on stopping illicit extraction, which signals that it sees this as an arms race rather than a problem one company's filters can solve alone.
