Stacy Garrity (official portrait)
Citing concerns about privacy and security, Pennsylvania Treasurer Stacy Garrity has banned the use of DeepSeek, a Chinese-owned artificial intelligence (AI) platform from all Treasury-issued devices.
“Our team [at Treasury] deals with billions of dollars that belong to the residents of the Commonwealth, so keeping our computer network secure is a top priority,” Garrity said.
The devices include laptops, cell phones, or other devices capable of connecting to the internet.
“There are growing fears that DeepSeek is directly linked to the Chinese Communist Party, potentially allowing the Chinese government to obtain sensitive government or personal data,” Garrity said. “Banning DeepSeek is necessary to ensure the safety of Pennsylvanians’ hard-earned tax dollars and other important, sensitive information entrusted to Treasury.”
DeepSeek was founded in July 2023 by High-Flyer, a hedge fund based in Hangzhou, Zhejiang, China and became the most downloaded app in the United States in late January, according to Covington Inside Government Contracts.
DeepSeek R1, its latest model released in January, rivals ChatGPT-maker OpenAI, while costing far less to create, per BBC.
YOU MAKE OUR WORK POSSIBLE.
The firewall for the state’s Treasury Department has also been updated as a result of the order to block access to the DeepSeek app and its corresponding website from its network. The system at the Treasury is “designed to block access to any new AI platform to allow necessary and thorough vetting.”
Sarah Rajtmajer, an Assistant Professor of Information Sciences and Technology at Penn State, told the Capital-Star she thinks Garrity’s reasons for banning the DeepSeek AI – due to concerns over privacy and security risks – are valid.
“I think that it is a Chinese company, and large language models in general, including DeepSeek, ingest information that you give it, which is often personal information, in order to feed the model right into performance task,” Rajtmajer said. “And so if you don’t trust the model…if they’ve decided that that’s not smart for security reasons, then I think that would be a reason to say, ‘Okay, we do not want to be handing over data that we don’t know what what individuals will be sharing.’”
Rajtmajer said people are using these large language models like DeepSeek and ChatGPT for a lot of things that are varied and creative, meaning anyone can type anything into those prompts.
“So I think it’s cautious, but probably smart, if you’re worried about where that data is going, to be careful,” she said.
Garrity isn’t the first elected official in the United States to ban DeepSeek as a result of security concerns. Over the last few weeks, governors in Iowa, Texas, New York, and Virginia issued directives banning DeepSeek from government state devices.
Researchers from Cisco and University of Pennsylvania’s Engineering program found DeepSeek’s AI model, R1, failed to block malicious prompts in security tests, exposing “major safety flaws.”
“Having a model that’s not safe, can have terrible [repercussions],” said Mahdi Sabbaghi, a PhD student in Electrical and Systems Engineering. “We need safeguards on using all of the elements, not only DeepSeek.
GET THE MORNING HEADLINES.
It remains to be seen if the United States federal government will take action banning DeepSeek from their devices, although U.S. Reps. Josh Gottheimer (D-NJ) and Darin LaHood (R-IL) announced earlier this month that they are introducing legislation to ban DeepSeek on government devices.
On Feb. 17, South Korea announced it was suspending new downloads of DeepSeek, and would resume service of the app after improvements were made in accordance with the country’s privacy law, per Reuters.
This is not the first time Garrity has directed her agency to ban a Chinese-owned app, on the basis of security concerns. In Dec. 2022, Garrity joined 16 other states at that time in announcing that TikTok, a social media app owned by Beijing-based company ByteDance, was banned from all Pennsylvania Treasury-issued devices.
“It’s a very similar concern as the Tiktok concern,” Rajtmajer told the Capital-Star. “Although the TikTok concern is not related to state information, it’s related to personal information of citizens in the country and data.”
Since DeepSeek is relatively new, Rajtmajer doesn’t think banning it on Treasury devices will have a big impact on the current day-to-day operations of the office.
“I think, and there are plenty of other large language models,” she added. “DeepSeek is not the only one, so I can’t imagine that this will be impactful in that way, aside from like social impacts or economic impacts possibly related to the signaling of this move, but from like a technical standpoint, DeepSeek is very powerful and very interesting to the technical community, because it is an open weight model. It is very exciting, technically speaking, but like just from day to day operations, I can’t imagine that it’s deeply embedded in what they’ve been doing because it’s so new.”
While this move was made by Pennsylvania’s Treasury Department, Rajtmajer also provided a warning to individuals who use any large language model, regardless of whether it’s DeepSeek, ChatGPT, or others that are similar, saying their data is at risk of being shared with a third party.
“So while it makes sense that the government has additional concerns about the nationality of the company, from the individual’s perspective, their privacy is just as at risk, regardless of whether the company is Deepseek or ChatGPT,” Rajtmajer told the Capital-Star. “And that from their perspective, they should use both with equal awareness of the personal information that they share.”