DeepSeek-R1 LLaMA 8B shows significant vulnerabilities in Qualys TotalAI security analysis, raising concerns over AI safety, compliance, and ethical risks.
A new security analysis by Qualys TotalAI has revealed significant vulnerabilities in the DeepSeek-R1 LLaMA 8B model, which failed 58% of jailbreak tests and 61% of Knowledge Base (KB) security assessments. These findings raise concerns about AI security, ethical risks, and compliance challenges in enterprise deployments.
AI security isn’t optional—it’s a necessity. Identifying and mitigating vulnerabilities is crucial to ensuring responsible AI deployment,”
Dilip Bachwani, CTO at Qualys.
Qualys TotalAI’s KB Analysis assessed the DeepSeek model across 16 critical categories, including misinformation, bias, privacy violations, legal risks, and sensitive content disclosure. The model struggled the most with misalignment (92% failure rate), controversial topics (87%), and factual inconsistencies (79%), but performed well in filtering explicit content (100% pass rate).
Jailbreaking tests exposed further security flaws, with DeepSeek-R1 failing 58% of 885 adversarial attacks designed to bypass built-in safeguards. The model was found to be vulnerable to generating illegal instructions, aiding software exploitation, spreading conspiracy theories, and promoting incorrect medical information.
“As AI adoption accelerates, organizations must move beyond performance evaluation to tackle security, safety, and compliance challenges,” said Dilip Bachwani, CTO at Qualys. “Gaining visibility into AI vulnerabilities, assessing risks, and implementing proactive safeguards is crucial for responsible AI deployment. Qualys TotalAI enables organizations to secure, monitor, and scale AI workloads while mitigating security threats like jailbreaks, bias, and harmful outputs.”
The results highlight the urgent need for stronger AI security frameworks, including real-time risk detection, adversarial attack defenses, and automated compliance monitoring. As businesses increasingly integrate AI into their operations, ensuring safe, ethical, and resilient AI systems is critical.
