AI Generated Image
Anthropic Uncovers 'Largest Known Distillation Attack'
Newser • • business
Key Points:
- Anthropic has accused Chinese tech giant Alibaba of covertly copying its proprietary AI technology through a technique called "distillation," involving unauthorized access via tens of thousands of accounts to generate millions of chatbot interactions.
- This practice, described as the "largest known distillation attack" on Anthropic, allegedly aims to train Alibaba's own AI models using Anthropic's systems without permission.
- While distillation is common for open-source AI, US companies like Anthropic prohibit it for their proprietary models and ban Chinese companies from using their AI, though loopholes allow some access, such as through corporate accounts at Chinese fintech firm Ant.
- Anthropic attempted to use hidden code to detect China-based users to prevent misuse but withdrew the measure due to privacy concerns, warning that such distillation attacks pose serious national security risks.
- In response, Alibaba announced it will block employees from using Anthropic's AI tools for work starting Friday, signaling increased caution amid these allegations.
