I’m reminded of that recent incident where it was discovered that the South African regulatory agency responsible for writing rules for the safe and appropriate use of AI had used an LLM to generate the text they submitted and apparently not validated it prior to the submission.
I don’t because that implies that the AI can lie. What I mean by that is it implies that the AI has the contextual understanding to say something that isn’t true for the purposes of deception. AI can’t do that.
When an AI LLM provides an answer that isn’t true, it’s a rounding error.
I’m reminded of that recent incident where it was discovered that the South African regulatory agency responsible for writing rules for the safe and appropriate use of AI had used an LLM to generate the text they submitted and apparently not validated it prior to the submission.
https://www.cnbcafrica.com/2026/south-africa-pulls-ai-policy-after-hallucinated-citations-expose-drafting-scandal
I think that we need to reframe the verbiage from ‘hallucination’ or ‘mistake’ to lies.
AI lies all the time. Confidently and to the harm of it’s users.
I don’t because that implies that the AI can lie. What I mean by that is it implies that the AI has the contextual understanding to say something that isn’t true for the purposes of deception. AI can’t do that.
When an AI LLM provides an answer that isn’t true, it’s a rounding error.