Microsoft investigating whether DeepSeek illegally used OpenAI's training data for their own model

Microsoft and OpenAI are investigating whether Chinese AI startup DeepSeek illegally accessed OpenAI’s data to train its own AI models. The allegation is that individuals linked to DeepSeek may have used OpenAI’s API to exfiltrate large amounts of data, as per a report by Bloomberg.

This API is the primary way that businesses and developers access OpenAI’s services. Microsoft’s security team had flagged suspicious activity last fall, regarding OpenAI’s services and the use of its API by certain entities. The tech giant, which is the largest investor in OpenAI, has since informed the company about the potential breach.

The news comes shortly after DeepSeek, known for its low-cost AI solutions, caused a stir in the tech world. The startup’s free AI assistant recently overtook OpenAI’s ChatGPT in downloads on the Apple App Store, leading to a significant selloff in tech stocks. This move has raised concerns about DeepSeek’s rapid rise and its implications for US AI companies. Microsoft’s investigation into DeepSeek’s actions could potentially reveal unauthorized use of proprietary data from OpenAI’s models, adding to the growing tension between Chinese and American AI firms.

More from Tech

OpenAI launches ChatGPT Gov for US govt agencies amid threat from Chinese AI companies Trump’s tariffs make rupee hit record low. What does this mean for the markets, economy?

White House weighs in on the allegations

David Sacks, the White House’s AI and crypto czar, spoke out about the situation, suggesting that DeepSeek might have stolen intellectual property from US companies. In an interview with Fox News, Sacks stated that there is “substantial evidence” to suggest DeepSeek could have “distilled the knowledge” from OpenAI’s models, potentially using stolen data to fuel their own AI advancements. This is a serious claim, as the unauthorized use of data could have broader implications for the US AI industry and national security.

Sacks’ comments reflect growing concern within the US government about China’s increasing efforts to replicate and compete with American AI technologies. The situation has sparked further debate about how to protect intellectual property in the fast-moving AI sector, particularly as foreign companies develop models that challenge US dominance.

OpenAI responds to data theft concerns

In response to the Bloomberg report, OpenAI issued a statement acknowledging that China-based companies, among others, are frequently trying to replicate US AI models. While the company did not specifically name DeepSeek, it confirmed that efforts to protect its intellectual property (IP) are ongoing. OpenAI emphasized that it engages in countermeasures to safeguard its data, including a careful process for determining which capabilities to release in public models.

OpenAI also stressed the importance of working closely with the US government to protect its most advanced AI models from being replicated or stolen. As the race to dominate AI technology continues, safeguarding intellectual property has become a top priority for US companies, especially with the growing competition from Chinese firms. With the DeepSeek investigation still ongoing, it remains to be seen what consequences, if any, will follow from these allegations of data theft.