Recently in the United States, the company OpenAI has disclosed evidence suggesting that the Chinese artificial intelligence startup DeepSeek has been utilizing OpenAI’s proprietary models to train its own. This revelation has raised concerns in the tech industry about potential intellectual property infringement.
On Monday, January 27th, the low-cost application DeepSeek surpassed OpenAI’s ChatGPT in downloads on the Apple App Store in the U.S., causing a sell-off in tech stocks.
According to reports from the U.K.’s Financial Times, OpenAI has noted what appears to be evidence of DeepSeek utilizing a technique known as “Distillation.”
“Distillation” is a method that allows developers to achieve better performance on smaller models by employing the outputs of larger, more powerful models, ultimately enabling them to obtain similar results for specific tasks at a lower cost.
While “Distillation” is a common practice within the industry, DeepSeek’s potential use of this technique to create its own models for competition with OpenAI may violate the service terms outlined by OpenAI. The company has declined to provide further details or evidence. OpenAI’s service terms prohibit users from “replicating” any of their services or “developing models that compete with OpenAI using output results.”
Additionally, according to Bloomberg’s January 28th report, it has been revealed by sources that Microsoft and OpenAI are investigating whether data output from OpenAI technology has been obtained without authorization by a group associated with DeepSeek. Last fall, Microsoft’s security researchers apparently discovered that individuals, suspected to be linked to DeepSeek, had been extracting large amounts of data using OpenAI application programming interfaces (APIs).
White House AI and Cryptocurrency Envoy David Sacks told Fox News on Tuesday that there is a potential for intellectual property theft to have occurred.
In its latest statement, OpenAI expressed awareness that Chinese companies and others are continuously attempting to refine models of leading U.S. AI companies. The company stated, “We have taken countermeasures to protect our intellectual property, including carefully deciding which cutting-edge capabilities should be incorporated into released models and believe… we are working closely with the U.S. government to best protect the most powerful models from the influence of adversaries and competitors attempting to acquire American technology, which is crucial.”
BBC reported Crystal van Oosterom, AI venture partner at OpenOcean, as commenting that “DeepSeek appears to be built on publicly available research foundations from major institutions and companies in the U.S. and Europe.” Nevertheless, the impact of building on the achievements of others remains unclear.
DeepSeek’s AI model R1, launched this week with its touted low-cost advantage, caused market turbulence. Claiming a cost advantage of 20 to 50 times, the R1 model positioned itself as a low-priced alternative to products from OpenAI and Meta. This news sparked a strong global market response, leading chip giant Nvidia to see a 17% plunge in its stock price on Monday, evaporating $593 billion in market value, marking the largest single-day loss and dragging down U.S. tech stocks across the board.
On Tuesday, January 28th, White House spokeswoman Karoline Leavitt announced that the U.S. National Security Council is reviewing the impact of DeepSeek on national security, calling it a “wake-up call for the U.S. artificial intelligence industry.”
Moreover, CNBC has learned that the U.S. Navy issued an email warning to sailors, explicitly banning the use of DeepSeek’s applications “due to potential safety and ethical concerns.” A Navy spokesperson confirmed the authenticity of the email.