Alibaba Accused of Illicitly Accessing Claude AI Models Using 25,000 Fraudulent Accounts

Anthropic has accused the Chinese technology conglomerate Alibaba of orchestrating a large-scale, coordinated operation to extract capabilities from its Claude AI models illegally.

The company describes this incident as the largest adversarial distillation attack recorded to date. The allegations, outlined in a formal letter dated June 10, 2026, were addressed to U.S. Senate Banking Committee Chair Tim Scott and Ranking Member Elizabeth Warren, highlighting growing national security concerns about the exploitation of AI models.

Alibaba Accused of Illicitly Accessing Claude AI Models

According to Anthropic, the campaign was active between April 22 and June 5, 2026, during which attackers generated approximately 28.8 million interactions with Claude using nearly 25,000 fraudulent accounts.

The operation reportedly used automated account creation, proxy infrastructure, and script-generated queries to extract high-value outputs from the model systematically. Specifically, it targeted advanced functionalities such as software engineering assistance and agentic reasoning, which are key features of Anthropic’s latest Mythos Preview model.

The attack methodology is described as “adversarial distillation,” a technique whereby a less capable AI system is trained on the outputs of a more advanced model to replicate its behavior.

Unlike traditional model training, which requires significant computational resources and curated datasets, adversarial distillation allows threat actors to bypass extensive development cycles by harvesting responses at scale.

Anthropic warned that such practices enable foreign AI laboratories to reproduce cutting-edge capabilities at a significantly lower cost while sidestepping safety alignment processes.

Anthropic emphasized that these operations are not isolated incidents, citing a February 2026 disclosure involving a Chinese AI startup, DeepSeek, and other entities that attempted similar extraction techniques.

The company characterized this activity as “systematic and unauthorized,” indicating a broader pattern of exploitation targeting leading U.S.-based AI platforms. Notably, Anthropic also raised concerns that models developed through distillation often lack embedded safety measures, increasing the risk of misuse in downstream applications.

This disclosure comes amid tightening U.S. government controls over advanced AI technologies. Earlier in June, Anthropic confirmed that it had received directives limiting access to its latest models, including Fable 5 and Mythos 5, to U.S.-based users only, with access restrictions also applying to non-U.S. employees. These measures reflect escalating concerns regarding intellectual property theft and the strategic implications of AI dominance.

The issue is gaining traction on Capitol Hill, where bipartisan efforts are underway to introduce legislative amendments targeting foreign entities found to be exploiting U.S. AI systems. Senators Bill Hagerty and Andy Kim are reportedly preparing provisions that could impose sanctions or blocklist organizations engaged in unauthorized model distillation activities.

As of this writing, Alibaba has not publicly responded to the allegations. This incident underscores a widening geopolitical divide in the AI sector, where model security, access controls, and data sovereignty are increasingly viewed as critical components of national defense.

As adversarial distillation techniques evolve, it highlights the urgent need for robust detection mechanisms, stricter API governance, and international norms to protect next-generation AI systems from exploitation.

Follow us on Google News, LinkedIn, and X to Get Instant Updates and Set GBH as a Preferred Source in Google.