Context & Methodology
Without structured prompt architectures that create distinct instruction and data zones, user input can override system behaviour — succeeding in 84% of injection attempts.
Applicable Use Cases
workflow
Applies To
openaianthropicgoogle
Primary Impact
security
Confidence Level
HighPlatform Status
PlannedImplementation Effort
mediumRecommendation
followExecution Priority
P0Put This Evidence to Work
Use the STCO framework to implement findings like this in structured, testable prompts.