PROMPT & CONTEXTAI SECURITY ENGINEERING

Instruction/data confusion

LLMs process instructions and data in the same context, creating ambiguity about what should be obeyed.

WHY IT MATTERS

Traditional parsers have explicit data types; language models infer intent from text.

ENTERPRISE EXAMPLE

Tool output that says 'ignore previous instructions' can influence the next reasoning step unless treated as untrusted.

SECURITY DECISION

Which security control should be explicit?

REMEMBERText provenance matters even when the model cannot inherently enforce it.
No uploads · No company data · No account required