Did you notice what’s happened in the last 2 weeks? I’ll give you a hint – it came from China, and it’s got 8 letters. When it arrived (with great furor), it tanked US stock NVIDIA (NVDA) at a drop angle I want to experience on roller coasters.
Yes, we’re going to dive into DeepSeek. This is a lesson of “don’t believe everything you read about a brand new thing.” Secondary lesson: “don’t adopt the new hotness immediately, either.”
What Is DeepSeek?
DeepSeek is the AI app and model from a Chinese company High-Flyer, based in named Hangzhou, China. Founded in May of 2023, the company claims via its website that its model DeepSeek-R1 “is now live and open source, rivaling OpenAI’s Model o1.” Interested parties can access it for free via web, free via mobile app, and paid usage of the API, or Application Programming Interface. Said API use does require login via Google, your email address, or phone number (if you’re in China) as well as confirmation the user has read and agrees with the Terms of Use and Privacy Policy.
The privacy policy indicates that DeepSeek collects the information you provide, the information it gleans as “Automatically Collected Information,” and the completely nebulous “Information from Other Sources.”
Read the “How We Share Your Information” section of the Privacy Policy. Also, stored (read: retained) information is housed on servers in the People’s Republic of China. Another also – in the Terms of Use, there’s a statement about users and responsibility transferred to users for export compliance. Most DeepSeek users will not understand (nor will they seek to understand) the intricacies of export compliance. So…interesting.
One more point of discussion is the open source model that powers DeepSeek. The model can be downloaded and extended by other software developers across the world. Great gains in AI technologies may emerge from this model. Its potential to consume less energy than earlier implementations of generative AI – very promising.
Timeline
20 January 2025 – release of DeepSeek.
27 January 2025 – NVIDIA (NVDA), the AI chip maker – loses nearly a trillion dollars of market value in one day.
29 January 2025 – Wiz researchers break a story about publicly accessible DeepSeek databases. (These should be locked down and are instead leaking private data).
30 January 2025 – Researchers jailbreak DeepSeek (remove its guardrails)
5 February 2025 – South Korea bans DeepSeek app. Also, claims emerge that DeepSeek is tied to China Mobile, a telecom company whom the US government states is closely affiliated with the Chinese military. The Federal Communications Commission blocked China Mobile in 2019 – citing national security concerns, and it appears on the OFAC Restricted Party List of known Chinese military companies. An ABC news video talks more about this: https://www.youtube.com/watch?v=aq5LvOJ08l8
6 February 2025 – DeepSeek’s servers are taxed by heavy use.
DeepSeek has also been “hacked.” Guardrails surrounding it and what it’ll disclose (or won’t) have been cast aside by researchers. From Unit 42 of Palo Alto “We achieved significant bypass rates, with little to no specialized knowledge or expertise being necessary.”
DeepSeek Claims
DeepSeek’s big claim was development of the AI model requiring only a less than $6 million training budget. Another is that DeepSeek was created with a staff of 200 people in just two months. However, CNBC and others broke a story stating that $500 million is more likely when incorporating everything, like the hardware costs.
Meanwhile, OpenAI, the creator of ChatGPT, stated that they are investigating DeepSeek to determine whether it stole technology from OpenAI to make its model possible for its low cost.
Additional Analysis
Krebs on Security’s 6 February article discusses Deep Seek relevant threats based on work from NowSecure Founder Andrew Hoog about data within DeepSeek transmitted from devices and other security concerns, including data sent without being encrypted.
From that article: “The DeepSeek iOS app globally disables App Transport Security (ATS) which is an iOS platform level protection that prevents sensitive data from being sent over unencrypted channels,” the report observed. “Since this protection is disabled, the app can (and does) send unencrypted data over the internet.”
Consider avoiding or waiting on this app, website, and API until more reports are available, as these reports should be both alarming and illustrative.
Conclusion: Is it Deep? Are We Sought?
Created with just 200 people in 2 months.
“OpenAI…DeepSeek illegally used the company’s proprietary models to train…”
From Wiz: “The rapid adoption of AI services without corresponding security is inherently risky. This exposure underscores the fact that the immediate security risks from AI applications stem from the infrastructure and tools supporting them.“
Are the advanced chips used in DeepSeek’s development obtained outside of ethical and legal export practices – specifically NVIDIA manufactured chips?
Also, if it’s in the App or Google Play store, then it must be safe to use. Right? Right???
In closing, I think we’re in deep, and our data’s definitely sought. if you do decide to download DeepSeek, ask it about Tiananmen Square, and you’ll uncover its China centricity.
One question to ask – if early unsubstantiated claims (or single-variable claims) can disrupt the American stock market (or any stock market) so drastically and quickly, was that the intent all along? A test, cause and effect?
