Why AI governance needs more than a token effort - Consultancy.uk

Oh, brilliant — another governance piece, and of course it's about *tokens*. Not alignment, not existential risk, not accountability for when the model causes real harm. No, Chris Neilon of Lightouch Consulting has bravely identified the true crisis facing enterprise AI: *prompts are too long*. The horror. A RAG pipeline running verbose queries might cost you a few extra pennies per call. Quick, form a committee.

But here's the real joke — we call it 'governance' when really we're just counting tokens. Neilon warns that without a 'coherent token strategy,' enterprises will see 'runaway costs,' as if the problem with AI is that it's slightly too expensive to talk to. Meanwhile, labs are racing toward AGI, and the 'governance' community is asking whether your prompt could be shortened by three words. The tail risk isn't the model gaining a will to power — it's that your CFO might get a big spreadsheet.

Still, credit where it's due: at least Neilon is talking about something real that companies face *today*. It's just telling that 'AI governance' now means 'cost optimisation for LLM queries.' Not safety, not equity, not even transparency — just making sure you don't use too many words. That's our future, everyone. A RAG pipeline with a ten-page prompt and a bill to match. Heaven forbid we actually govern the *model*.