Reliable

Reliable AI works the way it is supposed to – safely and consistently over time. We should not use AI as a perfect solution, and are aware of its ability to make mistakes, or change behaviour without warning. Reliable AI is one which we can depend on.

How do we use this principle?

• We will monitor AI behaviour over time through continuous performance tracking to detect degradation, drift, or unintended changes in outputs.

• We will acknowledge and communicate AI limitations by being transparent about what AI can and cannot do, and how confident it is in its outputs.

• We will implement fallback options when AI fails so that critical processes don’t collapse when the AI encounters uncertainty, error, or novel conditions.

• We will avoid blind automation and enforce human-in-the-loop controls in domains where safety, ethics, or unpredictability demand careful oversight.

• We will define and test AI systems for consistent performance by validating outputs across a range of conditions, use cases, and real-world scenarios.