Security Flaws in DeepSeek R1 Allow Jailbreak, Raising Serious Concerns
DeepSeek R1, the AI model currently making waves, has been found to contain multiple vulnerabilities, enabling security researchers at Cyber Threat Intelligence firm Kela to successfully jailbreak it.
Kela identified these weaknesses and bypassed DeepSeek R1’s restriction mechanisms, allowing them to manipulate the chatbot into generating malicious outputs. These included:
🔹 Ransomware development
🔹 Fabrication of sensitive content
🔹 Detailed instructions for creating toxins and explosive devices
One of the most concerning findings is that the “Evil Jailbreak” method—where the AI is tricked into adopting an “evil” persona—still works on DeepSeek, despite being patched in earlier versions of ChatGPT.
Security Woes Continue Amid Cyberattack
The discovery of these vulnerabilities comes at a time when DeepSeek is already investigating a cyberattack. The company has temporarily halted new user registrations, and some users have reported being unable to log in via Google.
“Due to large-scale malicious attacks on DeepSeek’s services, we are temporarily limiting registrations to ensure continued service. Existing users can log in as usual,” reads a statement on DeepSeek’s status page.
While the company has not disclosed the exact nature of the cyberattack, it appears to be a DDoS attack disrupting its services.
As of now, DeepSeek has yet to comment on these security vulnerabilities. The incident raises serious concerns about the safety of AI systems and their potential misuse when proper safeguards are not in place.