Alibaba Used 29 Million Fake Exchanges to Extract Claude's Intelligence, Claims Anthropic
Terming it “unauthorized,” the Dario Amodei-led startup has urged lawmakers to introduce stronger protections against AI model extraction.
News
- OPAQUE and TII Launch Open Standard for Verifiable AI Governance
- e& UAE Launches OneWork, a Unified Communications Platform Hosted Entirely in the UAE
- Alibaba Used 29 Million Fake Exchanges to Extract Claude's Intelligence, Claims Anthropic
- Anthropic's Mythos Finds Critical Flaws in Classified US Systems: Report
- Qatar Launches AI-Focused Scholarship Program to Build Future Digital Workforce
- China's LineShine Tops TOP500, Becomes World's Fastest Supercomputer
[Image: Chetan Jha/MITSMR Middle East]
American artificial intelligence research laboratory Anthropic has asserted that Chinese technology conglomerate Alibaba is conducting a widespread initiative to exploit the capabilities of its AI model, Claude, in an unethical manner.
In a letter to US Senators Tim Scott and Elizabeth Warren, Anthropic accused operators linked to Alibaba of carrying out about 29 million exchanges with its headliner model by creating thousands of fake accounts.
The campaign relied on “distillation attacks,” a machine-learning technique in which the output of one advanced model is used to train another. While model distillation is a recognized AI development model, Anthropic alleges that in this instance it was carried out without permission by creating fraudulent user accounts.
The Dario Amodei-led startup has urged lawmakers to introduce stronger protections, such as imposing penalties, against the extraction of AI models, terming it “unauthorized.” It argues that such operations enable competitors to leverage and replicate AI research and development at a fraction of the original cost.
Anthropic has described the operation as the largest extraction campaign of its kind.
This comes as competition between US and Chinese companies intensifies to dominate the AI race. Recently, China’s LineShine displaced America’s El Capitan from the top position in the latest rankings of the ‘Top500’— a feat last achieved after 2017.
Notably, the US Department of Defense has established links between Alibaba, BYD, and Baidu to the Chinese military. Concerns about distillation attacks aren’t new. Previously, rival OpenAI accused Chinese groups of using similar techniques to build competing AI models.
Anthropic and OpenAI are gearing up for a public listing later this year— one that will have everyone watching.
