Anthropic Rallies Industry to Combat AI Model Theft
Anthropic Rallies Industry to Combat AI Model Theft
Publish Date: 2026-02-23 17:34:00
Source Domain: www.pymnts.com
- Anthropic accuses three Chinese AI labs, DeepSeek, MiniMax and Moonshot AI, of illicitly using its AI model “Claude” to train their own models.
- The labs used a technique called “distillation,” which allows for quicker and cheaper acquisition of advanced capabilities.
- Anthropic identified 24,000 fraudulent accounts used by these labs to generate 16 million exchanges with its model, violating its terms of service and regional restrictions.
- Anthropic argues that such illicit distillation campaigns could result in models lacking necessary safeguards and threaten America’s technological leadership in AI.
- The company is taking action to combat these attacks by developing detection tools, sharing intelligence, strengthening access controls, and creating countermeasures.
- Anthropic emphasizes the urgency of coordinating efforts among industry players, policymakers, and the global AI community to address this growing threat.
- Google Threat Intelligence Group (GTIG) has also observed an increase in model extraction attacks, advising AI model providers to monitor API access for suspicious patterns.
