Anthropic claims that Alibaba's Qwen lab executed a major AI model theft involving 28.8 million queries. This operation reportedly utilized around 25,000 fake accounts to interact with Claude, targeting its advanced coding capabilities.
The alleged attack took place over six weeks, from April 22 to June 5, 2026. According to Anthropic, this tactic allowed the extraction of Claude's capabilities without triggering alarms, as the activity remained below standard rate limits.
Model distillation, a technique utilized in the alleged theft, involves feeding prompts to a strong AI model and using its responses to train a cheaper model. This method raises concerns about ethical practices in AI development.
A letter outlining these allegations was sent to US Senators Tim Scott and Elizabeth Warren on June 10, 2026, and was made public in late June. As of now, Alibaba has not responded, and no regulatory actions have been initiated.
If proven true, these actions could suggest that some of Qwen's recent advancements were attained not through innovation but rather through the unauthorized use of a competitor's proprietary technology. This incident may have implications for the broader AI market, potentially affecting investor confidence in security measures within AI firms.
This material is for informational purposes only and does not constitute financial advice.
