Generative AI at Work / Unit 2 GAI 046 worked example, 60 to 90 seconds
Technical, Misuse, Ecosystem
The idea
NIST's profile suggests one way to group its twelve risks, drawn in part from a United Kingdom scientific report, into three families that point to different owners and fixes.
In practice
Step one: technical risks come from the model malfunctioning: confabulation, dangerous recommendations, harmful bias and homogenization, the value chain, and data privacy. These belong with the people who choose and test tools. Step two: misuse risks come from people using the tools to do harm: weapons information, abusive content, information integrity, information security, human–AI configuration, and again data privacy. These call for policy, access control and monitoring. Step three: ecosystem risks play out across society: environmental impacts, intellectual property, and once more data privacy. These need organization-wide decisions, not fixes inside one team.
Where people slip
The grouping is NIST's suggestion, not a rule, and NIST notes that some risks cut across all three; data privacy appears in every family.
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