Summary:
- Microsoft's security researchers have discovered a new attack technique called "LLMjacking" that exploits large language models (LLMs) like ChatGPT to gain unauthorized access to systems and networks.
- The attack involves tricking the LLM into generating malicious code or content that can be used to compromise the target system, bypass security controls, and gain a foothold for further attacks.
- Microsoft has provided guidance on how to mitigate the risks of LLMjacking, including monitoring for suspicious activity, implementing robust access controls, and keeping LLMs and other AI systems up-to-date with the latest security patches.