AI and intellectual property: Karma returns for Open AI

Open AI, the creator of ChatGPT, told the Financial Times that Chinese AI system DeepSeek had illegally exploited its data to train its model. Distillation is a technique employed by AI model developers, in which the results generated by an advanced AI model are used to train a smaller model. The larger, more complex model is usually called the teacher model, while the smaller one is called the student model. The aim of distillation is to enable the student model to achieve a level of performance comparable to that of the teacher, while using fewer computing resources.

OpenAI's terms of service prohibit users from copying its services or using its results to "develop models that compete with OpenAI". In addition, the application is raising national security concerns at the White House, and its impact is currently under review by the National Security Council. White House spokeswoman Karoline Leavitt said, "This is a wake-up call for the U.S. AI industry."

David Sacks, named "AI Czar" by President Donald Trump, warned that over the next few months, leading US AI companies will start taking steps to "try to prevent distillation", which would "certainly slow down some of these copycat models". Companies like OpenAI can prevent third parties from using distillation by restricting their access: banning accounts, blocking IP addresses or limiting the number of queries third parties can make. However, it is not certain that these methods are effective in completely excluding distillation. OpenAI has announced that it is using "countermeasures" to protect its intellectual property, adding "as we move forward (...) it is extremely important that we work closely with the U.S. government to better protect the most successful models from the efforts of adversaries and competitors to appropriate U.S. technology."

It's particularly tasty to see that American AI companies, who shamelessly use copyrighted content to train their systems, are now showing such concern for protecting intellectual property. indeed, this was the point raised by Mike Masnick, founder of tech news site TechDirt, who commented on OpenAI's reaction to the distillation as follows: "I'm in the minority here on Bluesky believing that training AI systems is not copyright infringement. But seriously... for OpenAI to make that argument is really really ridiculous."