OpenAI is now requiring developers who want access to its cutting-edge AI models to verify their identities. The security enhancement is being linked to a Copyleaks study that pointed to striking similarities between China’s DeepSeek AI model and OpenAI’s work.
OpenAI is now requiring developers who want access to its cutting-edge AI models to verify their identities. The security enhancement is being linked to a Copyleaks study that pointed to striking similarities between China’s DeepSeek AI model and OpenAI’s work.
Officially, this step is aimed at combating abuse, but it is also intended to protect OpenAI’s work from being trained by competitors' artificial intelligence systems, Business Insider reports.
A new study by Copyleaks, a company specializing in AI content discovery, offers evidence of why OpenAI might be acting this way. Copyleaks used a system that identifies stylistic «fingerprints» of major AI models and calculated that 74% of the output from the Chinese model DeepSeek-R1 was interpreted as responses from OpenAI’s AI.
The Copyleaks method has also been tested on other models, including Microsoft’s phi-4 and Elon Musk’s xAI Grok-1. These models showed almost zero similarity to OpenAI — 99,3% and 100%, respectively, suggesting independent learning. Mistral’s Mixtral model has some similarity, but DeepSeek’s performance stands out sharply, which could indicate copying.
The study highlights that even when models are asked to write in different keys or formats, they still leave behind noticeable stylistic signatures—like linguistic fingerprints. These fingerprints are stored across tasks, topics, and prompts, and can now be traced back to their source with some degree of accuracy. This is of great importance for detecting unauthorized use of models, enforcing licensing agreements, and protecting intellectual property.
OpenAI previously said that Chinese companies are «constantly» trying to use American competitors to improve their AI models. OpenAI said it had found evidence of «distillation» that may have come from DeepSeek.
This method is used by developers to get better performance on simpler models by leveraging the results of larger, more powerful models. This allows them to achieve similar results at a much lower cost.
Recall that DeepSeek is a Chinese artificial intelligence lab funded primarily by High-Flyer Capital Management. It became popular in the winter of 2025 after its chatbot rose to the top of the Apple App Store charts.
DeepSeek’s language models, which were trained using efficient computational methods, have led many analysts and technologists in the West to wonder whether the US will be able to maintain its lead in the AI race and whether demand for AI chips will persist.
