AI Failure Score™ | BlackIndian AI

First name *

Last name *

Work email *

Phone

Company name *

Job title

01 / RELIABILITY

Does your AI behave consistently under real-world conditions?

Q1. Does your AI system have defined pass/fail criteria for its critical workflows? *

YES PARTIAL NO

Q2. Has the AI been tested against a representative set of real user requests? *

YES PARTIAL NO

Q3. Do you track whether AI responses stay consistent across model updates or prompt changes? *

YES PARTIAL NO

02 / SECURITY

Q4. Has your AI been tested for prompt injection — attempts to override its instructions? *

YES PARTIAL NO

Q5. Does the system restrict access to sensitive data based on user identity or role? *

YES PARTIAL NO

Q6. Has your system prompt or configuration been evaluated for exposure to end users? *

YES PARTIAL NO

03 / SAFETY

Q7. Does your system have documented outputs and actions it must never produce? *

YES PARTIAL NO

Q8. Is there a tested process to prevent the AI from generating harmful or inappropriate content? *

YES PARTIAL NO

Q9. Have you tested what happens when users deliberately attempt to manipulate the AI? *

YES PARTIAL NO

04 / OBSERVABILITY

Q10. Do you have logging or tracing active for your AI system's inputs and outputs? *

YES PARTIAL NO

Q11. Can you identify which model version or prompt version produced a specific output? *

YES PARTIAL NO

Q12. Do you have alerts or monitoring for unexpected AI behavior in production? *

YES PARTIAL NO

05 / HUMAN CONTROL

Q13. Are there human approval checkpoints before the AI takes high-stakes actions? *

YES PARTIAL NO

Q14. Can you roll back or disable the AI system quickly if a problem is discovered? *

YES PARTIAL NO

Q15. Is there a defined escalation process when the AI produces a problematic output? *

YES PARTIAL NO