OpenAI claimed it has found evidence suggesting that DeepSeek may have used its proprietary models to train an open-source competitor.
It suspects the Chinese start-up engaged in "distillation," a technique where outputs from larger AI models help optimize smaller ones.
While common in AI development, OpenAI contends that using this method to create a rival system violates its terms of service.
Last year, OpenAI and Microsoft investigated and blocked accounts believed to belong to DeepSeek for suspected unauthorised use of OpenAI’s API. DeepSeek did not respond to requests for comment.
DeepSeek’s AI advancements have drawn industry attention, particularly due to its cost-effective model training.
Some experts believe its latest AI model may have been trained on OpenAI’s GPT-4 outputs.
OpenAI continues efforts to safeguard its technology, warning that AI firms worldwide attempt to extract knowledge from leading US models. Meanwhile, OpenAI itself faces copyright lawsuits over its training practices, as many media companies, writers and other content producers have accused it of scraping their content.
"Chinese firms in intellectual property theft shock," mock-exclaimed analyst Neil Wilson at Tip Ranks, noting that several analysts this week have questioned some of the claims by DeepSeek around processing units and costs, "but this takes it a stage further".
He added: "it goes to the heart of the Trumpian claim to be at the start of a golden era for the US. That's hardly going to be case with China undercutting its technological lead.
"If OpenAI can show the White House enough evidence DeepSeek stole its IP, it's going to be grist for the mill for Trump's trade-tariff war agenda with China."