Deepseek, a Chinese AI startup that captures many of the recent topics of artificial intelligence (AI), stated that it has quoted malicious attacks and restricts registration for services.
“We have temporarily restricted registration to secure continuous services by large -scale malicious attacks on Deepseek’s services,” he said on the Insident Report page. “Existing users can log in as usual. Thank you for understanding and support.”
Users who are trying to sign up to an account say that registration may be a vision, stating that you need to wait and try again, and a similar message is displayed.
“It’s not a surprising surprise that DeepSeek is being targeted for malicious web traffic because of the growing popularity,” said Knowbe4’s security advocate, Eric Kron, stated in a statement shared with Hacker News. I am.
“This kind of attack may be a way to force an organization by stopping the attack and promises to recover availability. It is a rival organization that tries to adversely affect competition. I want to protect their investment by taking competition.
Deepseek, founded in 2023, is an emerging company in China, who “AGI (artificial general information) is dedicated to reality, according to the explanation of the facial face page.
The company has become a hot topic in the AI world, and the iOS chatbot app has reached the top free application chart in the UK and the United States this week, leaving Openai’s Chatgpt.
For the past month, AI Research Lab has released a series of inference and expert language models under Mit license, which claims to be better than Silicon Valley rivals. The results faced US sanctions that prohibit the sale of advanced AI chips to Chinese companies.
“To train Deepseek-V3 in each token before training, we need only 3. 7 days in a cluster with 180K H800 GPU time, 2048 H800 GPU,” said the company.
“As a result, the stage before the training is less than two months, it is 2664K GPU time. Context long expansion combines 119k GPU time and 5K GPU time after training, Deepseek-V3 only takes 2.788m GPU time. No, the rental price of H800GPU is only $ 5.576.
That said, this platform has been found to censors to delicate topics such as Tiananmen Square, Taiwan and China’s Uyghurs treatment.
At the end of last year, security researchers Johann Laverger revealed security defects in DeepSeek chatbots. This may have been abused by malicious actors and controlled the user’s account via a quick injection attack, including cross -site scripting (XSS) payload.
The threat intelligence company Kera is a report released on Monday, and despite being better than meta (lama) and mankind (class), the Deep Shek model is ethically violated by chatbots. He revealed that it is susceptible to evil jailbreak Persona attacks that can provide answers to questions. Or safety restrictions.
This includes ransomware development, content manufacturing, detailed instructions on the creation of toxins and explosive devices, and the generation of malicious outputs, such as cord snippets for Star Malware.
It is noted that the privacy policy also has the personal information of users, including devices and network connection information, use patterns, and payment details, is hosted on the “Safe Server in the People’s Republic of China”. For Washington in the ban on Tactoku.
However, China allows Internet companies around the world to operate in Japan as long as they follow local laws and regulations, and have never called for the government to collect or provide data located overseas. He states that he does not ask companies or individuals. For local laws.
“We live on a timeline where non -US companies continue to live on Openai’s original mission.
Openai’s CEO Sam Altman called Deepseek’s R1 Reasoning Model “impressive” and said that having a new competitors is legal.
