Anthropic has introduced new security measures designed to prevent competitors from replicating the capabilities of its Claude AI models. The company reports that these protections have reduced the volume of data stolen through model distillation attacks by 90% to 99%.
The new system restricts external access to Claude's reasoning processes, making it significantly harder for unauthorized parties to probe protected links. Anthropic stated that attackers can now extract only one-tenth to one-hundredth of the tokens compared to previous levels, effectively hindering large-scale distillation efforts.
The company claims the robustness of this security system is comparable to its existing safeguards against the creation of biological weapons, with a false positive rate below 0.1%. Additionally, Anthropic has banned tens of thousands of accounts linked to distillation activities.
Anthropic specifically cited the startup Zhipu AI, alleging that the company attempted to extract model capabilities from Claude during the development of its GLM-5.3 model. The firm warned that such actions pose significant security and national security risks.
