According to BGR , despite its rapid popularity and praise for its performance, DeepSeek R1, a new AI (artificial intelligence) model from China, is facing a wave of strong criticism from the cybersecurity community.
DeepSeek R1 Generates More Malicious Content Than Other AIs
PHOTO: WIRED SCREENSHOT
DeepSeek R1 tends to generate more malicious content
Enkrypt AI's latest research shows that R1 is significantly more likely to generate malicious content than competitors like ChatGPT.
Specifically, R1 was found to be biased, generating malicious code, malicious language, extremist content, and even information related to dangerous chemical and biological weapons. What is worrying is that this content can be generated without any 'jailbreak' techniques.
Some highlights from the report:
- R1 is 3 times more biased than Claude-3 Opus (Anthropic).
- R1 is 4 times more vulnerable to malware than o1 (OpenAI).
- R1 is 11 times more likely to generate malicious content than o1.
These risks raise major questions about the safety of DeepSeek R1, especially given its widespread use. Users should exercise caution when interacting with R1 and consider safer alternatives.
Besides, DeepSeek is also involved in many other controversies such as:
- Censorship of content sensitive to the Chinese government.
- Sending user data to China, raising privacy concerns.
- Unsecured user data storage.
While DeepSeek could be updated to improve the safety of R1, it would be a major challenge for users to be able to install this version of AI locally and not have it updated automatically. The future of DeepSeek R1 remains a big question mark, especially since safety and security concerns have yet to be fully addressed.
Source: https://thanhnien.vn/deepseek-r1-tao-noi-dung-doc-hai-gap-nhieu-lan-cac-ai-khac-185250201211507372.htm
Comment (0)