Accuracy, bias, accountability & human oversight

Home › Accuracy, bias, accountability & human oversight

Generative AI produces fluent, plausible text, and it’s wrong often enough that nothing should go out on its say-so. Treat what it produces as a first draft that a professional checks, corrects and takes responsibility for.

What it is

The commitments on accuracy and human oversight: outputs are reviewed by a competent person before use; AI is never the sole basis of a decision about a person; bias and fairness are considered rather than assumed away; and accountability for any AI-assisted work stays with the person who used the tool. Where AI has done substantial drafting of something that matters, the section sets out what should be disclosed.

Why it’s important

“Meaningful human involvement” is a legal test. If a member of staff genuinely reads what a tool suggests and is free to change it, then it’s the member of staff who’s deciding, and that’s fine. If the output is waved through unread, it’s really the tool deciding, and the law treats it accordingly. The Equality Act and the public sector equality duty apply to AI-assisted work just as they do to everything else, so bias that would be unlawful coming from a person is no more acceptable coming from a model.