UK cyber agency warns LLMs will always be vulnerable to prompt injection

The UK National Cyber Security Centre warns that prompt injection is an intrinsic, persistent security weakness in large language models (LLMs): because models do not distinguish instructions from data, malicious prompts can bypass guardrails and be used to jailbreak, exfiltrate data, or even enable remote code execution in integrated workflows (e.g., CI systems, browser agents). The report outlines attack examples, explains why the issue may never be fully eliminated due to model architecture, and notes industry efforts (monitoring, evaluation changes) to mitigate but not eradicate these risks.