SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-direct-prompt-injection

What weakness does a direct prompt injection exploit?

Combining untrusted input with higher-trust instructions.

Start playing free

  1. Separating data and commands
  2. Concatenating untrusted input with instructions ✔
  3. Removing all user input

Why: A language model sees instructions and user text in one context. Malicious user text can compete with application instructions and induce unintended behavior.

Huh, didn't know that: Prompt injection can target privacy, integrity, or availability.

The dad joke
The prompt invited untrusted text, which immediately rearranged the furniture.
Like this card?
Play the game

References

Answer: Concatenating untrusted input with instructions. A language model sees instructions and user text in one context. Malicious user text can compete with application instructions and induce unintended behavior.

  1. NIST Adversarial Machine Learning Taxonomy nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-2e2025.pdf
    A PROMPT INJECTION overrides these instructions, exploiting the concatenation of untrusted user output to the system prompt
    Checked 2026-10-10.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceCan one prompt filter fully prevent every injection attack?Artificial IntelligenceWhere does an indirect prompt injection hide?Artificial IntelligenceHow do AI, machine learning and deep learning fit together?

That's stuff everyone should know.