Attacks on AI systems · Question 12 of 12
Why is prompt injection considered hard to fix completely, unlike SQL injection?
1 answer is correct. Hard
Show answer
Answer
- A Models do not strictly separate commands from data
Explanation
With SQL injection, prepared statements reliably separate code and data. A language model, by contrast, processes everything as text in the same context. That is why several layers are used: filters, minimal rights, confirmations and monitoring.