US-based artificial intelligence (AI) company Anthropic has accused Chinese AI companies DeepSeek, Moonshot, and MiniMax of illicitly extracting the company’s Claude AI agent’s capabilities to improve their own AI models, as per an official blog post.
Anthropic alleges that the above-mentioned Chinese AI outfits created around 24,000 fake accounts to generate more than 16 million exchanges with Claude, using a technique known as ‘distillation’ to train their own AI models.
To explain, distillation is a process by which an inferior AI model acquires the capabilities of a relatively superior model by training on the latter’s outputs. The California-based AI company remarks that distillation is a ‘legitimate training method’, but that it can be used for illicit purposes.
Anthropic says that competitors can employ distillation to acquire powerful capabilities from other AI models in a fraction of the time and cost that it would take to develop these independently and organically.
What Did Anthropic Find?
The AI firm found that MiniMax ran the greatest effort, producing more than 13 million interactions, while Moonshot generated over 3.4 million, and DeepSeek focused on extracting step-by-step reasoning traces.
Importantly, Anthropic said that all these operations targeted Claude’s strongest differentiators rather than general consumer queries. And in MiniMax’s case, the AI company even detected activity in real time before the distilled model was released.
Why Is Anthropic Worried About Distillation?
The San Francisco-headquartered AI company remarks that illicitly distilled AI models are found wanting when it comes to adequate safeguards and safety valves.
“(AI) models built through illicit distillation are unlikely to retain those safeguards, meaning that dangerous capabilities can proliferate with many protections stripped out entirely,” Anthropic warns.
The AI company further argues that it – along with other US-based AI outfits – prevents state and non-state actors from utilising AI to build bioweapons or conduct ‘malicious cyber activities’
It also warns that overseas labs which distil American models can feed these unprotected capabilities into military, intelligence, and surveillance systems, thereby enabling authoritarian governments to deploy frontier AI for offensive cyber operations, disinformation campaigns, as well as mass surveillance.
What Is Anthropic Doing About This?
The Dario Amodei-led AI company says that it continues to ‘invest heavily’ in defences that make such distillation attacks harder to execute and easier to identify. These include:
- Detection: Anthropic has built several classifiers and behavioural fingerprinting systems designed to identify distillation attack patterns in API traffic. This includes the detection of chain-of-thought elicitation used to construct reasoning training data.
- Intelligence sharing: Anthropic says that…
Source link
Disclaimer
We strive to uphold the highest ethical standards in all of our reporting and coverage. We blogs.grocliq.com want to be transparent with our readers about any potential conflicts of interest that may arise in our work. It’s possible that some of the investors we feature may have connections to other businesses, including competitors or companies we write about. However, we want to assure our readers that this will not have any impact on the integrity or impartiality of our reporting. We are committed to delivering accurate, unbiased news and information to our audience, and we will continue to uphold our ethics and principles in all of our work. Thank you for your trust and support.
Website Upgradation is going on for any glitch kindly connect at [email protected]
