Anthropic Suspended Hong Kong Accounts Over Model Theft
The firm blocked users in restricted regions for allegedly using Claude to train competing AI models.
Updated on Oct. 2, 2026 in Artificial Intelligence
Live Poll
Should global technology companies restrict service access based on a user's geographic location?
Anthropic has suspended user accounts in Hong Kong, Macau, and mainland China after detecting unauthorized access through proxy networks. The company alleges that these accounts were used to perform industrial-scale model distillation to train competing AI systems.
Why it matters
This enforcement action highlights the intensifying conflict over intellectual property in the AI sector, as firms scramble to prevent competitors from scraping proprietary model outputs to build rival services. It marks a significant escalation in how AI developers are policing regional access to protect their core technology.
Anthropic reported that Moonshot AI routed 23 million exchanges to its Claude service between May and July. The company identified seven Chinese AI labs involved in these activities, which were facilitated by VPNs routed through Singapore and fraudulent payment registrations.
The players
Anthropic
A San Francisco-based AI research company focused on the development of Claude, a family of large language models designed for safety and steerability.
Moonshot AI
A Beijing-based artificial intelligence firm specializing in long-context window models and natural language processing applications.
The details
The unauthorized access relied on virtual private networks (VPNs) to spoof location data by masking connections through Singapore or Japan. Anthropic claims these accounts were part of an industrial effort to extract Claude's responses—a process known as model distillation—to refine and train separate, competing foundational models. The Beijing government has rejected these allegations of organized intellectual property theft.
Timeline
Between May and July, Moonshot AI allegedly routed 23 million exchanges to the Claude platform.
Anthropic published a formal report regarding industrial model distillation in September 2026.
Users reported widespread account suspensions and access denials starting October 1, 2026.
The Tech Race
This enforcement marks a pivot from passive monitoring to active disruption in the ongoing battle against model distillation. As labs attempt to iterate faster by harvesting the intelligence of competitors, the race to secure proprietary outputs has become as critical as the race to build the models themselves.
Users in Hong Kong, Macau, and mainland China will lose access to the Claude interface and API services regardless of whether they have active subscriptions. The company is actively auditing account sign-up data to identify and block further attempts to circumvent geographic access restrictions.
The takeaway
The incident underscores that developers will increasingly use usage-policy enforcement as a shield against intellectual property theft. Observers should track subsequent disclosures from Anthropic regarding the other six accused labs and any potential response from the Chinese AI sector.
Further reading
Learn more about the latest industry security challenges in Artificial Intelligence.
Live Poll
Should global technology companies restrict service access based on a user's geographic location?
