Generative AI is one of the hottest topics around, with hours of fun to be had from asking ChatGPT to summarize nuclear fission in the style of Chaucer, or Mid journey to depict a sports stadium inhabited by lemurs in the style of van Gogh. But how can we put it to work to really generate value? And how do we make sure that we can mitigate these technologies’ well-reported issues (information leakage, hallucination, bias, privacy violations…)? I will describe how we are putting Generative AI to work at Shell, and what we’re doing to ensure that it’s used safely.