- OpenAI accused Chinese startup DeepSeek of misusing its AI technology via distillation techniques
- Distillation involves smaller AI models learning from larger models by mimicking their responses
- The technique is legal but disputed if used to extract data from competitors without authorisation
Artificial intelligence (AI) has become the latest source of US-China tensions, with OpenAI accusing Chinese startup DeepSeek of misusing its technology.
The company told the House Select Committee on China that DeepSeek allegedly relied on a technique known as "distillation" to extract responses from advanced US AI systems and use them to train its own chatbot, R1," according to a memo obtained by Reuters.
The American AI giant stated that the Chinese firm was finding clever ways to bypass safety systems and trying to take advantage of the technology that US companies spent billions of dollars developing.
What Is Distillation And How Does It Work?
In the AI world, distillation is a common technique where a smaller or newer AI model learns by studying the responses of a larger, more advanced model. Instead of training that model completely from scratch, the newer model observes and mimics the advanced model's answers and behaviours.
Large AI models are expensive to run as they need powerful chips and a lot of computing power. So, by using distillation, companies can keep the quality of the large model by running a smaller, cheaper and faster version.
"By transferring the knowledge from a large pre-trained model to a smaller, more efficient model, distillation offers a practical solution to the challenges of deploying large models, such as high costs and complexity," analysts told The Forbes.
"This technique not only reduces model size and operational costs but also enhances the performance of student models for specific tasks," they said.
Is Distillation Illegal?
Distillation is widely used in the tech industry, especially when companies want to deploy AI on smartphones, laptops, or other devices that cannot run huge models. Distillation itself is not illegal. It is a common and accepted practice when companies use their own models.
But the dispute arises when a company extracts large amounts of AI outputs of other companies without authorisation and uses that data to directly compete with the original creator.
How Are Chinese AI firms Allegedly Stealing From American Firms?
Large American AI companies are spending billions of dollars to train their powerful models. Some Chinese AI firms, their US counterparts allege, are sending a large number of automated queries to American AI systems, collecting their responses and storing those outputs as training material.
Instead of building a model entirely from raw data, they use the answers from an already-trained system as learning material. OpenAI has claimed that it detected "obfuscated methods" such as rotating IP addresses and using proxy servers to avoid detection.
OpenAI claims that even though it has put protections in place to prevent people from misusing its models, some users linked to DeepSeek may have found ways around them. The company claimed that DeepSeek employees are using third-party routers to hide their identity. It further states that DeepSeek employees have developed code to access US AI models programmatically.
US Representative John Moolenaar is accusing China of following a repeated strategy when it comes to technology competition. "This is part of the (Chinese Communist Party) CCP's playbook: steal, copy, and kill," he said.
"Chinese companies will continue to distil and exploit American AI models to their advantage, just like when they ripped off OpenAI to build DeepSeek," he added.
Republican Representative Michael McCaul said Chinese companies are taking advantage of US technology. He referred to the American company Nvidia, noting that China reportedly built very advanced open-source AI models using less powerful Nvidia chips.
"DeepSeek should have been a wake-up call about the dangers of selling advanced semiconductor chips to the CCP. Using less powerful Nvidia chips, China developed the most advanced open-source models on the plane. I shudder to think of what they might do with more advanced hardware like the H200 chips," he said.
