COSAI SHARED RESPONSIBILITY FRAMEWORK

Test whether the CoSAI framework covers your scenario

Select a real-world AI incident scenario, build a five-layer accountability prompt, and paste it into your own chat or model. The site never calls an LLM for you.

This tool builds an analysis prompt in your browser. Nothing is sent to our servers for model inference. Use Copy prompt, then paste into ChatGPT, Claude, or any model you choose (that step uses your account, not ours). Do not paste confidential deployment details into a third-party chat unless your policy allows it. Preset cards stay on this page until you build a prompt.

Step 1: Select a scenario

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…

Step 2b: Analysis prompt

Copy this prompt and paste it into your own chat or model. The site does not run the model. Optional: show a deterministic local layer sketch (no LLM, no API key).

Step 3: Local layer sketch

Local sketch (no LLM)

Step 4: Framework scorecard

WHAT WORKS

    GAPS FOUND

      RECOMMENDATIONS

        Overall verdict

        Step 5: How other frameworks compare

        Where other frameworks fit

        Editorial: In our review, the CoSAI SRF is the only framework we found that assigns accountability at each technical layer: model providers, platforms, applications, data governance, and business use (unverified against all frameworks). Frameworks marked ADDS ENFORCEMENT provide legal liability at the legal-entity level (who gets fined); they work alongside SRF, not instead of it.