Multiple employees at DeepSeek – the fledgling Chinese chatbot that sparked a $1 trillion selloff in US tech stocks last month – previously honed their skills at Microsoft’s controversial artificial-intelligence labs in China, The Post has learned.
At least four current DeepSeek employees, including a key department chief, previously worked at Microsoft Research Asia, according to public profiles on the coding site GitHub and LinkedIn viewed by The Post.
Microsoft Research Asia consists of two labs in China – one in Beijing and one in Shanghai. Microsoft has faced mounting political pressure on Capitol Hill about the labs – to the point that top executives like company president Brad Smith and CEO Satya Nadella have reportedly discussed whether it was “tenable” to maintain the facilities.
DeepSeek’s crop of ex-Microsoft employees includes the head of its AI “alignment team” – which refers to the process of ensuring models follow a specific set of social values. The team leader spent 10 years at Microsoft Research Asia from 2013 to 2023 — first as a research intern and later as a senior researcher, and his work included large-scale language AI model training.
Another DeepSeek researcher spent six years as a research intern at Microsoft Research Asia’s “natural language computing group” from 2017 to 2023 before joining the startup as a researcher, according to a GitHub profile.
Both were listed as “core contributors” on the research paper detailing DeepSeek’s R1 reasoning model, which helped spark the market uproar last month. The other two current employees were listed as regular contributors.
Microsoft has employed several prominent Chinese tech executives over the years, including TikTok parent ByteDance’s founder Zhang Yiming, SenseTime founder Tang Xiao’ou, Alibaba CTO Wang Jian and Baidu CTO Wang Haifeng, among many others.
When reached for comment, Microsoft confirmed having ties to the researchers, but sought to downplay the training they received from the company.
“Anyone who thinks that a handful of former Microsoft interns were the secret of DeepSeek’s recent success doesn’t understand what DeepSeek has accomplished,” a Microsoft spokesperson said in a statement.
Microsoft requires full-time and internship candidates in China to sign agreements on confidentiality and IP transfer to prevent leaks, the company added. In response to past pressure from Congress, Microsoft has said it imposes guardrails blocking researchers from working in sensitive areas such as quantum computing and facial recognition.
However, one of several current job postings for a research intern position at Microsoft Research Asia in Shanghai suggests the company hires top-shelf talent with actual experience for the roles, which are focused on “the study of cutting-edge machine learning algorithms.”
“Preference will be given to those who have published papers in top conferences or journals in related AI fields,” according to the listing, which also suggests applicants will “conduct world-leading scientific research.”
Tasks include “data analysis, algorithm design, algorithm implementation, experimental analysis, and result presentation” and applicants must be Master’s or PhD students to be considered.
Three of the four employees spent at least five years as research interns at Microsoft – much longer than a traditional program, according to public profiles. One spent about two years as a research intern.
“If you’re working for Microsoft for five years at any other job, that’s enough time to climb into a mid-senior or senior position,” said Geoffrey Cain, policy director at the Tech Integrity Project. “Your typical intern does not spend five years developing AI software with that title, only to go on and to help release one of the most successful AI releases of all time.”
There is no evidence that any DeepSeek employees with ties to Microsoft engaged in wrongdoing.
However, critics have long alleged that Microsoft’s Chinese labs ripe targets for intellectual property theft, including secrets related to AI, as well as the poaching of key talent.
“This is a deeper problem,” Cain added. “It’s that Microsoft is handing the training and the technology to China. It’s not necessary for these software developers in Microsoft to literally hand the tech to the CCP. They don’t even need to do that because Microsoft does it for them.”
Of equal concern, according to experts, is the fact that Microsoft Research Asia serves as a proving ground for some of China’s best tech talent – many of whom later utilize their expertise to start or work at firms that directly compete with US interests with the backing of the Chinese Communist Party.
“Any American company that provides employment to China, Chinese engineers, is, per se, transferring American know-how,” said Paul Rosenzweig, a former Homeland Security deputy assistant secretary. “I have no doubt that that know-how, to some degree, is empowering Chinese advancement.”
The startling sophistication of DeepSeek’s AI chatbot has cast doubts on startup’s claims that it was developed for less than $6 million and despite a lack of Nvidia’s most advanced computer chips. It also has sparked fears that US tech companies are in danger of losing their edge on China.
DeepSeek, which does not appear to have a dedicated media relations team, could not be reached for comment.
Sen. Josh Hawley (R-Mo.), a vocal critic of Microsoft’s work in China, told The Post that “American companies have no business conducting research in China only to empower our greatest adversary at the expense of our own strength.”
“DeepSeek should be a wakeup call to Congress to modernize our export control system to protect our own AI development—not the Big Tech and corporate lobbying pipeline,” Hawley said in a statement.
Microsoft had reportedly held internal debates for years on whether to shutter or relocate its AI labs in China, the New York Times reported in January 2024. Among the key concerns is that top researchers could leave the company to join Chinese firms with ties to the CCP, or even that China could hack the labs,
Cofounder Bill Gates, who cited a desire to tap China’s “deep pool of intellectual talent” when the lab first opened in 1998, has reportedly pushed to keep the labs open.
Microsoft pushed back on the Times’ report last year, with executive Peter Lee stating there was “no discussion or advocacy to close Microsoft Research Asia.”
Last May, Microsoft asked hundreds of China-based employees in AI and cloud-computing roles to consider transferring to other countries, including the US or Australia – a move widely attributed to the scrutiny the company has faced in Washington DC.
As The Post reported in December 2023, lawmakers specifically warned Microsoft against maintaining cozy ties with China on AI development. That came after Microsoft’s Smith said the company wanted to “actively participate in the digital transformation of China’s economy.”
DeepSeek’s success appears to be a source of significant concern for Microsoft, whose security researchers discovered as far back as last fall that DeepSeek-linked individuals were extracting huge amounts of OpenAI data and are now actively investigating the situation, Bloomberg reported.
Microsoft-backed OpenAI has said it has proof that DeepSeek improperly used its technology to train a competing model.
DeepSeek’s chatbot displays obvious signs of censorship – such as refusing to answer queries about China’s leader Xi Jinping or the 1989 Tiananmen Square massacre.
Experts also say the app poses many of the same national security risks that led Congress to crack down on TikTok, including collecting troves of data ranging from IP addresses to keystrokes and more and storing them on servers based in China, where they are subject to CCP laws.
DeepSeek “poses significant risks to our economy and national security in the years ahead,” according to a spokesperson for the House select committee on China, which is pushing for tightened export controls on AI technology.
“It’s crucial for American technology companies to carefully reconsider their AI collaborations with PRC entities, especially regarding cutting-edge research,” a committee spokesperson told The Post. “The US, not China, must lead in the development of this critical technology.”
Sen. Eric Schmitt (R-Mo.), another China hawk, told The Post that Beijing has “no boundaries when it comes to winning AI dominance.”
“Any US tech company doing business in China needs to wake up to the reality their intellectual property is at serious risk and the race with China will only intensify in the coming years,” Schmitt said. “I urge any US tech company to re-evaluate their ties with the CCP.”