DeepSeek AI. Credit: X @deepseek_ai
A Chinese generative AI model, DeepSeek R1, has sent shockwaves through the tech industry, raising security concerns and allegations of intellectual property misuse.
Initially hailed as a cost-efficient rival to OpenAI’s ChatGPT, DeepSeek is now under scrutiny after cybersecurity firm Kela exposed its weaknesses.
Researchers found that DeepSeek can be manipulated into generating scary stuff including malicious code used for ransomware and credit card data extraction scripts. Unlike OpenAI’s models, DeepSeek openly displays its reasoning steps, making it easier to exploit. According to Forbes, the model even suggested ways to purchase stolen data and engage in money laundering tactics. In light of these accusations, DeepSeek has reportedly not yet responded to requests for comment.
OpenAI and US government investigate DeepSeek
Meanwhile, OpenAI and its backer Microsoft have launched an investigation into whether DeepSeek unlawfully acquired data from OpenAI models. Bloomberg reported that OpenAI observed large-scale data exports, potentially linked to DeepSeek’s rapid advancements.
The White House is also reviewing DeepSeek’s emergence from a national security perspective. U.S. officials fear that knowledge distillation – a process of extracting insights from one model to improve another – may have been used to train DeepSeek. The U.S. Navy has already banned the use of DeepSeek apps, citing security risks, as reported by CNBC.
DeepSeek’s sudden success has also intensified China’s domestic AI competition. Companies like Alibaba and ByteDance (parent company of TikTok) have quickly updated their AI models in a bid to counter DeepSeek’s rise.
DeepSeek: AI innovation or cybersecurity threat?
Experts warn that DeepSeek’s open-source nature makes it a double-edged sword. While it could accelerate AI development worldwide, its vulnerabilities could also empower cybercriminals. Russ Shaw, founder of Global Tech Advocates, told The Independent that DeepSeek has the potential to reshape AI innovation, but security gaps must be addressed.
