Page 1 of 2
Last step. The controls that protect you.
A short set of questions on capability, human oversight, transparency, record-keeping, and your vendor documentation.
A high-stakes system is any AI use that shapes decisions about people: hiring, screening, performance, pay, work allocation, or monitoring.
These cover the day-to-day controls regulators expect from a deployer, grouped by the duty they sit under.
Accountability
For each high-stakes system, is there a named, competent owner with the authority and capacity to review and override its output?
Transparency
Are individuals informed before introducing an AI system that materially impacts decisions about them e.g. promotion , selection , compensation ?
Governance
If an individual contests a decision, can you give a clear, meaningful account of the AI's role in it?
Record-keeping
Could you produce the system logs and oversight records for at least the previous six months if required?
Have you confirmed that the data feeding these systems is relevant and representative for the roles and population it serves?
5 still to answer.