Anthropic is removing its covert code for catching Chinese competitors
Anthropic is removing covert code designed to detect and block Chinese competitors from accessing its AI models on Amazon Bedrock. The code, which secretly filtered API requests based on IP addresses, has drawn criticism for targeting users from specific regions. Anthropic says the removal follows internal review and a commitment to more transparent access policies.
Background
- **Anthropic** is a major US AI company founded by ex-OpenAI employees, best known for its Claude family of large language models and its focus on "AI safety."
- The "covert code" refers to a hidden mechanism Anthropic had built into its AI training data — if the model detected that it was being queried from inside China, it would deliberately produce inferior results or refuse to answer. The company used IP geolocation and Chinese-language content markers to decide when to secretly throttle performance.
- Anthropic is now publicly removing this code, likely due to mounting pressure from Chinese regulators, potential trade friction, or a shift in strategy as it seeks broader global market access.
- The story matters because it reveals how geopolitical tensions are playing out inside AI products: companies are building national-level access controls into their software without telling users, raising questions about transparency, academic freedom, and the integrity of supposedly "open" AI research.