National AI Centre

Build in human override points. Make sure you have clear intervention points where humans can pause, override, roll back or shut down AI systems if needed.

Track, document and report relevant information about serious incidents and possible corrective measures to relevant regulators and/or the public in a reasonable timeframe. Reporting near-misses and corrective measures is good practice.

Unlike traditional software that follows explicit instructions, AI systems learn patterns from data and make their own opaque decision logic. This could mean automated monitoring for low-stakes applications, and mandatory human review for high-stakes...

For use cases requiring enhanced practices and GPAI systems, conduct independent (internal or external) and thorough review of testing and evaluation methodologies and results. Document and report any issues to the accountable person for the system.