DeepSeek, the Chinese AI startup that has captured much of the artificial intelligence ( AI ) buzz in recent days, said it’s restricting registrations on the service, citing malicious attacks.
” According to large-scale harmful attacks on DeepSeek’s providers, we are partially limiting licenses to ensure continued service”, the organization in an incident report page. Existing users can log in as normal. Thanks for your knowledge and support”.
Similar messages are displayed to people who try to for an account, warning them that registration may be occupied and that they should wait and try again.
It’s not surprising that DeepSeek is becoming targeted by malignant web traffic, according to Erich Kron, a safety awareness activist at KnowBe4, in a statement released by The Hacker News.
These kinds of attacks could be used to steal a company by promising to cease attacks and make available them for a fee, or to target rival businesses that want to harm the competition, or even individuals who have invested in a rival company and want to safeguard their investment by removing the competition.
The “large-scale malicious attacks” could be a distributed denial-of-service ( DDoS ) attack, according to Stuart Millar, principal AI engineer at Rapid7, where threat actors are reconfiguring their responses to sensitive questions to match those of other models or repeatedly attempting to jailbreak the system to extract the system prompt.
This might be attempting to maintain that state, such as removing the censorship that appears to exist in some topics, by deliberately changing how the model operates and trying to keep that state in place, Millar said.
” If one manages to extract what is known as the system fast, one of the biggest headaches for LLM providers is. This collection of original kick-off guidelines, if there is one that exists in DeepSeek, is likely to contain information on what to do, what not to do, links to other applications, and other information that may reveal more about the designers ‘ intentions.
DeepSeek, founded in 2023, is a Chinese upstart that’s “dedicated to making AG I]artificial general intelligence ] a reality”, according to a on its Hugging Face page.
The company has become the talking point in the AI world, with its iOS robot game the top of Apple’s Best Free Apps table in the U. K. and the U. S. this week, dethroning OpenAI’s ChatGPT.
In response to that forbid the price of superior AI chips to Chinese companies, the AI research facility has released a number of logic and mixture-of-experts language models under an MIT license that it claims is surpass its Silicon Valley rivals while being trained for a fraction of the cost.
” During the pre-training level, training DeepSeek-V3 on each trillion tokens requires only 180K H800 GPU time, i. e., 3.7 days on our swarm with 2048 H800 GPUs”, the organization in a study.
” Thus, our pre-training stage is completed in less than two weeks and costs 2664K GPU time. Combined with 119K GPU time for the environment size improvement and 5K GPU hours for post-training, DeepSeek-V3 costs just 2.788M GPU hours for its whole training. Assuming the rental price of the H800 GPU is$ 2 per GPU hour, our total training costs amount to only$ 5.576M”.
Despite this, it has been discovered that the platform censors responses to sensitive topics like Tiananmen Square, Taiwan, and the treatment of Uyghurs in China. This can be resolved by downloading and running the models offline locally.
Security researcher Johann Rehberger made the discovery of a security flaw in DeepSeek’s chatbot late last year that could have been used by a malicious actor to take control of a user’s account through a prompt injection attack involving an XSS payload.
In a released on Monday, Threat Intelligence revealed that are susceptible to evil jailbreak persona attacks that allow the chatbot to respond to questions that otherwise violate ethical or safety restrictions despite those from Meta ( Llama ) and Anthropic ( Claude ).
This included generating malicious outputs, such as ransomware development, fabricating content, detailed instructions for creating toxins and explosive devices, and code snippets for stealer malware.
In addition, according to its privacy policy, users ‘ personal information, including device and network connection information, usage patterns, and payment information, are hosted on” secure servers located in the People’s Republic of China,” which is likely to raise new issues for Washington in light of the TikTok ban.
China has asserted that it permits internet businesses from other countries to operate there as long as they adhere to local laws and regulations, and that it has never and will never ask for any company or individual to collect or provide data outside of China in violation of local laws.
” We are residing in a non-U.S. timeline. S.company is continuing the original vision of OpenAI: to conduct truly open, frontier research that empowers all, according to Jim Fan, senior research manager and lead of Embodied AI ( GEAR Lab ) at NVIDIA.
OpenAI’s CEO Sam Altman DeepSeek’s R1 reasoning model “impressive” and that it’s “legit invigorating to have a new competitor”.
Update
According to a from CNBC, the U.S. Navy has advised its members to refrain from using DeepSeek’s artificial intelligence technology, citing “potential security and ethical concerns related to the model’s origin and use.”
Italian data protection authorities have also been contacted by DeepSeek to inquire about the nature of the data that its web platform and mobile app collects. The Chinese startup has 20 days to respond.
This includes what personal data are collected, from which sources, for what purposes, what is the legal basis of the processing, and if they are stored on servers located in China, the Garante , adding it has also sought details on what kinds of information are used to train its AI models, and if data is gathered via scraping activities, and clarify how registered and non-registered users are informed about the processing of their information.
Found this article interesting? To read more exclusive content we post, follow us on and Twitter.