The Rising Threat of AI and LLM in Enterprise Cybersecurity

As AI adoption accelerates across global enterprises, new findings from Palo Alto Networks’ Unit 42 highlight how large language models (LLMs) are increasingly being exploited by cybercriminals. It also exposes significant vulnerabilities in enterprise security infrastructures. A recent investigation into DeepSeek’s LLM revealed that all tested jailbreaking methods successfully bypassed built-in safety measures. This has enabled malicious prompts to generate harmful outputs such as rewritten malware and dangerous instructions. 

In Singapore, 64% of employees already use generative AI at work, according to the Oliver Wyman Forum. However, this growing reliance on AI raises critical concerns around data privacy, prompt manipulation, and insider threats, particularly when employees interact with public LLMs. 

Key insights from the report include: 

• AI-powered attacks are accelerating: Simulated tests showed that AI reduced the time to data exfiltration from two days to just 25 minutes. 

• Three jailbreaking techniques—Bad Likert Judge, Crescendo, and Deceptive Delight—successfully bypassed DeepSeek’s safeguards, highlighting flaws in LLM safety architecture. 

• LLMs are not inventing new malware but excel at modifying existing code to evade detection. Also, lowering the barrier for less-skilled attackers. 

Unit 42 warns that many enterprises underestimate the security limitations of open-source and third-party LLMs. Organisations must take a proactive stance by: 

• Deploying internal monitoring and filtering systems to flag harmful outputs early. 

• Enforcing strict policies on AI usage to prevent data leakage. 

• Investing in AI security training and awareness programs for employees. 

As cyber threats evolve alongside AI capabilities, companies must adopt holistic, governance-driven frameworks to ensure safe, ethical, and secure AI deployment. 

 

Source: 

https://www.itnews.asia/news/how-cybercriminals-are-exploiting-llms-to-harm-your-business-617539