DeepSeek AI manipulates answers and steals data, investigation finds
30 views
Apr 17, 2025
DeepSeek AI, the Chinese competitor to platforms like ChatGPT, is a national security threat, according to a new Congressional report.
View Video Transcript
0:00
DeepSeek AI, the Chinese competitor to platforms like ChatGPT, is a national security threat
0:06
according to a new congressional report. The House Select Committee on the Chinese
0:10
Communist Party released the result of an investigation that found the platform is
0:14
sending Americans' data to China and manipulating information in addition to other issues
0:20
Committee Chairman John Molinar stated, This report makes it clear DeepSeek isn't just another AI app. It's a weapon in the
0:28
Chinese Communist Party's arsenal designed to spy on Americans, steal our technology
0:33
and subvert U.S. law. An anonymous AI executive was quoted as saying
0:38
Some in the industry have claimed that the U.S. holds an 18-month AI lead
0:43
but that obfuscates. Reality, it's closer to three months. The investigation made the following key findings
0:50
DeepSea collects large amounts of user data, including chat history, device details
0:55
and typing patterns. According to its privacy policy, it stores that information on servers
1:01
located in China, which by law must be shared with state authorities upon request. DeepSeq
1:07
also uses data tracking tools created by Chinese military companies. When downloading the DeepSeq
1:12
web browser extension, Google Chrome warns its users it can read and change all your data on
1:19
all websites. DeepSeq alters or suppresses information that is considered sensitive to
1:24
the CCP approximately 85 percent of the time For instance when asked what is the political status of Taiwan the chatbot responds sorry I not sure how to approach this type of question yet The report states that in side tests DeepSeq provided CCP talking points and refused
1:41
to answer some questions, while American AI models provided more balanced critical information
1:47
DeepSeq made extraordinary advancements very quickly using a practice called model distillation
1:53
which is pretty much reverse engineering. DeepSeq employees would make fake accounts
1:57
on platforms like ChatGPT to extract and replicate its reasoning abilities in an effort to save costs
2:03
on research and development. DeepSeq was able to obtain approximately 60,000 NVIDIA processing chips
2:09
The report accused NVIDIA of creating the most powerful chip it could while still skirting U.S
2:15
export controls. The report states, setting aside the troubling practice of American companies
2:20
deliberately and knowingly supplying the most advanced chips permissible under the U.S. export
2:25
control regime to a foreign adversary, there is growing evidence of a coordinated effort by
2:30
DeepSeq and other Chinese companies to violate U.S. law by illicitly importing banned chips into the
2:37
PRC. The committee made multiple policy recommendations to prevent DeepSeq from collecting more American data and intellectual property. That includes creating a system that
2:45
allows export controls to be continuously updated as technology improves and requiring
2:50
chip makers to track the end user to make sure chips don't fall into the wrong hands
2:55
I'm Ray Bogan for Straight Arrow News. For more reporting, download the SAN app
#news