AIResearchAIResearch
Machine Learning

Alibaba Allegedly Used 25,000 Fake Accounts to Steal Claude Capabilities in Largest AI Distillation Attack

Anthropic accuses Alibaba of using 25,000 fake accounts and 28.8 million interactions to illegally distill capabilities from Claude's advanced models.

4 min read
Alibaba Allegedly Used 25,000 Fake Accounts to Steal Claude Capabilities in Largest AI Distillation Attack

TL;DR

Anthropic accuses Alibaba of using 25,000 fake accounts and 28.8 million interactions to illegally distill capabilities from Claude's advanced models.

Em June 25, 2026, a notícia choca o campo de inteligência artificial ao revelar que a Alibaba utilizou uma campanha massiva, composta por 25.000 contas falsas, para tentar roubar as capacidades do modelo Claude da Anthropic. Essa operação, que ocorreu entre 22 e 5 de junho, destaca a crescente vulnerabilidade dos sistemas de IA de ponta diante de técnicas de extração industrial. A divulgação destaca a necessidade urgente de fortalecer controles e regulamentações para evitar abusos nessa área em expansão.

Conforme dados compartilhados por fontes independentes, a extracção de dados do Claude comprometeu o maior ataque já documentado à segurança de modelos de IA, com aproximadamente 28,8 milhões de interações geradas por contas fraudulentas. Essa estratégia não apenas expõe a tecnologia da Anthropic, mas também levanta questões sobre a eficácia das medidas de proteção atuais. A colaboração entre empresas de tecnologia e pesquisadores é essencial para garantir que esses avanços não sejam explorados de forma irresponsável.

O texto também aponta para um ponto importante, conforme destacado por um site especializado em análise de modelos de linguagem, que observa um aumento significativo na frequência de lançamentos de novos modelos. Essa tendência reforça a importância de manter a transparência e a segurança nesse cenário em constante evolução. Ações como as mencionadas indicam que a comunidade de segurança e os especialistas estão se adaptando rapidamente aos desafios que surgem.

O relatório indica que, em junho de 2026, a Alibaba foi acusada de empregar centenas de contas falsas para tentar replicar as habilidades avançadas do modelo Claude da Anthropic, representando uma das operações mais extensas já registradas em seu domínio blockonomi.com.

Essa atividade destaca um novo desafio de segurança que se contrapõe às tendências de abertura que a Anthropic promove, pois o uso de bots para acessar recursos de modelos de alto desempenho pode limitar a exclusividade de certas tecnologias pricepertoken.com.

Além disso, o caso reflete uma mudança estratégica nas relações industriais, já que empresas como a A24 começam a se envolver em parcerias com grandes laboratórios de IA, como a Google, para integrar inovações em suas operações criativas, apesar dos riscos associados à colaboração com atores já envolvidos em disputas legais sobre direitos autorais.

blockonomi.com report that on June 25, 2026, Alibaba allegedly deployed 25,000 fraudulent accounts to sabotage Claude's AI models. pricepertoken.com counters cite similar claims, emphasizing the pattern of misuse during this period. Both sources highlight the urgency to clarify intentions behind such actions.

The discrepancy between reports underscores challenges in verifying intentions without direct evidence. Contradictions persist, complicating consensus on whether these efforts align with stated goals.

Such activities necessitate stricter oversight to balance innovation with accountability, demanding coordinated global efforts.

---
aireleasetracker.com contextualizes the issue within broader AI advancements, noting how historical precedents inform current scrutiny.

Scale of the AllegedAI Distillation Campaign
Alibaba allegedly created 25,000 counterfeit user accounts between April 22 and June 5, 2026 Blockonomi. Those accounts generated roughly 28.8 million prompts that extracted Claude’s advanced reasoning and long‑task handling capabilities. The campaign spanned 44 days and represented the largest known distillation effort targeting a commercial AI system. Anthropic reported the breach to US Senators Elizabeth Warren and Tim Scott, prompting a request for stronger regulatory measures. The Commerce Department subsequently imposed global restrictions on Anthropic’s Mythos and Fable models after the incident.

The scale of the alleged operation raises concerns about competitive equity in the rapidly expanding large language model market. Analysts note that academic studies have documented smaller‑scale distillation attacks, yet this case involves a corporate actor with substantial resources. The partnership between Google and A24 demonstrates how major tech firms are positioning AI capabilities across creative and cloud domains. Regulators confront a challenge in defining illegal model replication versus legitimate fine‑tuning when proprietary data are involved. The episode will accelerate the development of stricter licensing frameworks and monitoring mechanisms for AI model usage.

Alibaba is accused of running a distillation attack that harvested roughly 28.8 million outputs from Anthropic’s Claude using about 25,000 fabricated user accounts. The campaign ran from April 22 to June 5, 2026, and targeted the model’s most advanced reasoning abilities. Anthropic says the stolen data was used to train cheaper models for Alibaba’s own AI products. The incident has already prompted regulatory scrutiny and restrictions on Anthropic’s newer releases.

The breach illustrates how API access can become a weapon for intellectual property theft in the AI era. It signals that future competition may hinge as much on defensive safeguards as on technical breakthroughs. Companies will need to rethink how they expose model capabilities to the public. Will the industry adopt stricter authentication standards before the next wave of espionage unfolds?

Frequently Asked Questions

How does AI model distillation work?
It involves querying a powerful model and using its responses to train a smaller, cheaper model.

What legal steps have been taken against Alibaba?
Anthropic reported the breach to US senators and the Commerce Department imposed restrictions on related models.

Why are fake accounts a problem for AI platforms?
They allow attackers to massively query models and extract proprietary outputs without detection.

Can regulators prevent API abuse like this?
Regulators are exploring new rules, but enforcement remains challenging across borders.

Will this affect the cost of using AI services?
Potentially, as providers may raise prices to fund stronger security measures.

About the Author

Guilherme A.

Guilherme A.

Former dentist (MD) from Brazil, 41 years old, husband, and AI enthusiast. In 2020, he transitioned from a decade-long career in dentistry to pursue his passion for technology, entrepreneurship, and helping others grow.

Connect on LinkedIn