CoSAI Wizards
CoSAI Stress Test
Back to wizards
COSAI SHARED RESPONSIBILITY FRAMEWORK

Is the CoSAI framework
actually functional?

Select a real-world AI incident scenario, run it through the five-layer framework, and see where accountability holds and where it breaks down.

01 — Select a scenario
💬
Chatbot PII leak
Customer service bot exposes user data via prompt injection
🤖
Autonomous agent
L4 agent makes unauthorized financial transactions
⚖️
Model bias in hiring
AI screening tool produces discriminatory outcomes
👤
Shadow AI
Employees use unapproved AI tools with sensitive data
⛓️
Supply chain attack
Compromised foundation model affects downstream apps

Example: A generative AI medical diagnosis tool running on Azure gives a patient an incorrect treatment recommendation due to hallucination. The hospital, the platform provider, and the model vendor all disclaim responsibility…

5/5 runs available

Analysis complete

WHAT WORKS

    GAPS FOUND

      RECOMMENDATIONS

        Overall verdict

        Is there a better model?