Microsoft Probes Whether DeepSeek Stole Data From OpenAI

Microsoft security team investigating whether individuals possibly linked to DeepSeek exfiltrated large amount of data from OpenAI

3 min
OpenAI logo displayed on a smartphone. Artificial intelligence, AI, ChatGPT
Getting your Trinity Audio player ready...

OpenAI and its main backer Microsoft are investigating whether Chinese AI start-up DeepSeek accessed OpenAI data in violation of the company’s terms and conditions to help build the Chinese firm’s innovative chatbot, Bloomberg reported.

Microsoft’s security researchers last autumn observed individuals they believe may be linked to DeepSeek exfiltrating a large amount of data using the OpenAI application programming interface (API), the report said, citing unnamed sources.

Microsoft notified OpenAI of the activity, the report said.

The activity could indicate that the people used unauthorised means to get around OpenAI’s restrictions on how much data they were allowed to access, the report said.

OpenAI chief executive Sam Altman at the OpenAI Hackathon in March 2018. Image credit: OpenAI
OpenAI chief executive Sam Altman. Image credit: OpenAI

‘Copycat’

OpenAI said in a statement that companies in China and elsewhere were “constantly trying to distil the models of leading US AI companies”.

It added that going forward it was “critically important that we are working closely with the US government to best protect the most capable models”.

“We are aware of and reviewing indications that DeepSeek may have inappropriately distilled our models, and will share information as we know more,” the company stated.

DeepSeek and its main backer, hedge fund High-Flyer, did not issue a comment during the week-long Lunar New Year holiday in China, which began on Tuesday.

White House AI chief David Sacks told Fox News that there was “substantial evidence” that DeepSeek had used a technique called “distillation” to create a “copycat model”.

“There’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI’s models,” Sacks said.

The White House has said officials are investigating the national security implications of DeepSeek’s model, which the one-year-old start-up said was developed at a fraction of the cost of better-known US models, while delivering comparable or better performance.

Intellectual property

The Chinese app rose to the top of Apple’s download charts in the West over the weekend, sending stock markets into a $1 trillion (£800bn) sell-off on Monday.

DeepSeek has acknowledged building upon publicly available research from European and American organisations, and its models are themselves open source.

AI firms themselves have been widely accused of violating intellectual property laws by allegedly using large amounts of copyrighted data to train their commercial models, with OpenAI targeted by multiple lawsuits over the issue.

Meanwhile, the US Navy has banned staff from using DeepSeek over security concerns.

Naval authorities emailed staff warning them not to use the app due to “potential security and ethical concerns associated with the model’s origin and usage”, CNBC reported.

Advertising