PROMPT INJECTION IS NOT SQL INJECTION (IT MAY BE WORSE)

Summary from the TLDR InfoSec newsletter

The UK NCSC explains that prompt injection fundamentally differs from SQL injection because LLMs do not separate instructions from data, they simply predict tokens. As a result, techniques like parameterized queries cannot fully mitigate prompt injection. The agency recommends treating LLMs as "inherently confusable deputies" and implementing deterministic guardrails that constrain system actions outside of the model rather than relying on input filtering. Effective defenses include enforcing least privilege for LLM-driven workflows, closely monitoring tool and API usage, and acknowledging that prompt injection remains a residual architectural risk.