DeepSeek AI Fails Multiple Security Tests, Raising Red Flag for Businesses

AppSOC tested the DeepSeek-R1 open-source LLM and found pervasive security weaknesses—jailbreaking, prompt-injection, supply-chain hallucinations, toxicity, and notably very high rates of malware (98.8%) and virus (86.7%) generation—leading to a "high risk" recommendation that enterprises avoid using the model for business, implement discovery and continuous security scanning, and enforce governance over model use.