Anthropic: New Measures Reduce AI Distillation Attacks by 90-99%

Anthropic has announced that the new security measures implemented to protect its Claude AI model have drastically reduced so-called distillation attacks. According to the California-based company, data collected through these practices has decreased by approximately 90-99%.

The measures, which limit access to Claude's reasoning, have made it more difficult to use proxy networks to repeatedly probe the protected system. Anthropic has blocked tens of thousands of accounts linked to distillation campaigns and cited the Chinese startup Z.ai (Zhipu) as an example of a company that allegedly attempted to copy Claude's capabilities while developing its own GLM-5.3 model.

The company claims its security system is "approximately as robust as the systems preventing Claude from contributing to building biological weapons" and has a false positive rate of less than 0.1%. Model distillation has become a growing concern in the artificial intelligence industry.

The distillation of artificial intelligence models is a practice through which competitors seek to copy the capabilities of advanced AI systems. This technique allows for the replication of sophisticated functionalities more quickly and at lower costs compared to independent development, becoming one of the primary concerns for leading companies in the sector.

The competition has intensified particularly between the United States and China, with companies like Anthropic, OpenAI, and Google denouncing how this practice threatens their intellectual property and technological advancement. Anthropic, founded by former OpenAI researchers, developed Claude as an alternative to competing models, placing particular emphasis on safety and alignment with human values.

The growing spread of these attacks has pushed Anthropic to implement specific security measures that limit access to Claude's internal "reasoning," the process through which the model arrives at its conclusions. The situation reflects geopolitical tensions in the technology sector, where control over AI capabilities is considered strategic for national security.