Generative AI at Work / Unit 5 GAI 117 worked example, 60 to 90 seconds
Checking Against Policy
The idea
The M-CaReM-AI method's first defence is cross-checking AI output against the real policies and data it is supposed to reflect, not against what such policies usually say.
In practice
Step one: identify which policy, procedure or rule each statement in the output depends on. Step two: open the current, approved version of that policy, and confirm the version and its date. Step three: compare each rule in the output with the policy's own wording, including conditions and exceptions. Step four: flag anything that sounds like general best practice rather than your organization's rule, and confirm it with the policy owner.
Where people slip
Models tend to generalize from typical documentation. A rule that is true almost everywhere can still be wrong for your organization, and that is the error a policy check exists to catch.
Your place on the trail
Every lesson rests on earlier ones and opens later ones. Tap any to jump.
Something wrong in this lesson? Flag it · Sources for Generative AI at Work