Research question
Can separating prompt and data channels at the interface and in training prevent injection?
Key findings
- A secure front-end plus structured instruction tuning reduces attack success to near zero for optimisation-free attacks - Optimisation-based attacks remain partly effective
