A recent letter sent to US Senators has revealed that Anthropic, the developer of the Claude AI model, has accused Alibaba of orchestrating the largest attack yet in an attempt to clone its model. The alleged attack, which occurred between April 22 and June 5, involved the creation of nearly 25,000 fraudulent accounts that generated over 28.8 million exchanges with Claude. Anthropic claims that Alibaba's AI lab, Alibaba Qwen, used obfuscation techniques and proxy networks to evade detection and targeted some of Claude's most valuable capabilities, including agentic reasoning and software engineering. This incident highlights the growing concern over the security of AI models and the potential for malicious actors to exploit them. The attack is also seen as an attempt by Alibaba to bypass the significant training and R&D costs required to develop its own frontier model. Anthropic warns that such attacks have become widespread and could undermine the significant investments made in AI research and development in the US. The incident has also sparked concerns over the potential for a 'circumvention economy' to emerge, where malicious actors develop and share techniques to evade detection and exploit AI models. The allegations against Alibaba come after similar attacks were reported by OpenAI and Google, highlighting the need for increased vigilance and cooperation to protect AI models from such threats.

Source: Ars Technica