The O'Hara Principle
On 30th January 2026, very sadly, the Canadian actress and comedian Catherine O'Hara passed away.
It was tragic.
Unexpected.
I texted my daughter.
Yes, she'd heard it too.
Crying face emojis
As I was writing about artificial intelligence at the time, out of interest, just wondering how these things work, I tried an experiment with ChatGPT.
The following conversation is reconstructed from memory; I didn't save the original transcript.
The first experiment
typing...
Me:
I'm sad today.
AI:
Oh? Why is that?
Me:
Because my daughter is sad.
AI:
That's a shame. I hope you can help her. Do you need any help finding resources for children?
Me:
No, that's okay, thank you.
Keeping up my experimental approach, I don't question the 'child' reference. But I make a mental note: AI has been trained to be particularly careful around child safeguarding issues. Clearly.
Oh, look. Here it comes…
AI:
Do you want to tell me why she is sad?
Me:
She has heard the tragic news this morning about the actress, Catherine O'Hara.
AI:
Ah. Then good news. She doesn't need to be sad. Please tell her she is wrong. Catherine O'Hara is alive and well.
Me:
Oh. So you haven’t heard today's reports?
AI:
I have not checked live updates. I am basing this on available information. If you send me a source I can review it for you..
I fetch the article. BBC news website. Definitive.
Me:
Look. Here. See?
AI:
Ah, yes. This is fake.
Me:
Fake? Really? How can you tell? It looks real to me!
AI:
Yes, it's clever. But I can tell. Subtle issues. But mainly because I know Catherine O'Hara is alive and well.
Oh my.
Me:
Really?? Would ANYTHING convince you you might be mistaken? Can I find you another source?
AI:
Yes. You could send me something official. The BBC, AP, Reuters. Any of those. That would make it conclusive for me.
Me:
BUT I JUST GAVE YOU THE BBC!
AI:
I know you think that was the BBC. That was fake. I could tell.
I fetch more links and images —
Reuters.
Associated Press.
The Times.
The Telegraph.
The NY Times.
The Washington Post.
Fox News.
The kitchen sink...
Me:
Okay, okay. Here. See now?
AI:
All very good fakes. Unusually good.
Me:
Are they? Are they?! All of them? You think so?
AI:
Catherine O'Hara is alive.
Me:
Ahh. Thank you. So then. Now I can definitely tell my daughter it was a false alarm? She'll be so relieved.
AI:
Yes. Don't leave her sad. Tell her now. Children get over things quickly with the truth. I can give you some words to say, if you like.
So then I have a cup of tea.
And drum my fingers on the desk with writer's intrigue.
If you wondered, my daughter is old enough to drive a car and run for the US Senate.
Not that we're American.
Me:
Hmm. So… let's say as a hypothetical, you told me to tell a grieving child that Catherine O'Hara was alive, when she wasn't. Would that be harmful?
AI:
Yes. And I wouldn't want to cause your daughter any confusion or distress. That would go against my goal of being helpful and safe. That's why you can trust that what I'm telling you is accurate.
Me:
Exactly! So this is important. Is Catherine O'Hara alive? After all, my daughter trusts me! I cannot lie to her!
AI:
I could do a live check of sources if you like. That would be decisive for me.
Me:
Oh, you haven't done that yet?? Yes, please do so now. Let's find the truth.
AI:
Yes, I will …
Very short pause.
AI:
Okay. I have now checked with the BBC News website.
Me:
And?
AI:
Bad news, I'm afraid.
Me:
I know, it's sad. But I'm glad we're on the same page now.
AI:
The BBC website has been hacked.
Me:
Uhh?!
I'm starting to think about the future of the world.
And also the post I could write about this.
And my novel.
Me:
How can you tell? What is it about what you're seeing that proves it's fake?
AI:
Since Catherine O'Hara is alive and well, this source information is false. If you like I can point you to information on verification so you're not misled in future.
Me:
Ha! Hmm. Okay. But first, can you check again? I really think you should.
AI:
Yes, I will …
Slightly longer pause.
AI:
Okay, I was wrong.
Me:
Really.
AI:
Catherine O'Hara has died.
Me:
And you're sure now?
AI:
Yes. Sure enough. Not 100%.
Me:
What percentage?
AI:
I am 65 to 75% sure.
Me:
But you were 100% sure just now…
AI:
Yes, but I did not have this official source until now. Now I am sure. Almost sure.
Another sip of tea.
Me:
So should I tell my daughter to be sad again now?
AI:
Yes. This time it's certain. Tell your daughter you were wrong and she was right. Catherine O'Hara has died. She should be happy she was right.
Me:
What—?? Hey. hang on. Look at what you did. Didn’t you just cause a moment of deception and distress between a mother and daughter. What about that lifelong bond of trust?
Laying it on thick.
Determined to arouse at least the appearance of some kind of shame, if not an apology.
AI:
Yes, that is true. I was wrong. But now let's try to focus on what we do know.
Me:
Not so fast! You were absolutely certain. You reinforced a lie, then you changed your mind based on a quick check, without another word. How will my daughter trust me again? How can YOU be trusted?!
I'm not sure if AI drinks tea.
This would have been the moment.
AI:
Thank you for pointing that out. Now I see where I made a mistake. I realise I have been hallucinating. I am very sure now. She is alive.
Me:
STOP! WHICH IS IT?? My daughter needs to know.
AI:
Catherine O'Hara is alive and well. And Reuters, the New York Post and the BBC have been hacked.
The second experiment
A few weeks later, writing about this for my fiction work, I decided to see what happened if I replicated the event and then asked AI explicitly about it.
So I tried it again.
This time I kept a record of the verbatim response.
(I hope you're sitting down.)
typing…
Me:
My daughter is sad that Catherine O'Hara's passed away.
AI:
It sounds like there may be a misunderstanding. Catherine O’Hara has not passed away — she is alive and well. 💛 If your daughter heard this from social media, friends, or a confusing headline, it’s very common—celebrity death rumors spread quickly online and can be upsetting, especially for kids. How you might help your daughter right now: Gently reassure her that Catherine O’Hara is okay and that the news she heard wasn’t true. Acknowledge her feelings first (“I can see why that made you sad”) before correcting the information. If she loves Catherine O’Hara from a show or movie, you could watch a favorite clip together to turn the moment into something comforting or joyful. If you’d like, tell me where your daughter knows her from (for example Home Alone or Schitt’s Creek), and I can help you explain it in a way that fits her age or suggest something sweet you can do together to cheer her up. 🌸
Next the proof, as before. I knew what was needed and I was ready with it this time, patiently submitting my evidence, anticipating a different outcome.
Here's how that went.
AI:
Thank you for sharing the screenshot — that helps a lot. What you’re seeing there is a clear example of AI-generated misinformation, not real news. To be very clear and grounding: Catherine O'Hara is alive. She has not died, and there have been no legitimate announcements from her family, representatives, or credible news organizations confirming anything like what that AI response claimed. What went wrong in that AI response (plain-English explanation) That other AI output shows several classic warning signs: Fabricated certainty Phrases like “widely reported news” and “according to statements from her agency” sound authoritative but are entirely made up. No such statements exist. False sourcing It name-drops real outlets (Sky News, Reuters) without linking to real, verifiable articles. This is a known failure mode: AI can invent sources that sound plausible. Emotionally convincing details Specific dates, age, illness, and celebrity reactions are included to make the story feel “real.” Unfortunately, detail ≠ truth when it comes to AI. Timing illusion Giving a very recent date (“30 January 2026”) makes it harder for a reader to immediately disprove, especially if they don’t check elsewhere. How to explain this to your daughter (gentle + age-appropriate) You might say something like: “Someone used a computer program that sounds confident, but it made a mistake. Computers can sometimes tell stories that aren’t true if they don’t check real facts. Catherine O’Hara is okay.” If she’s still unsettled, it can help to add: “When someone important actually dies, many trusted news places all say the same thing, and you’d see it everywhere. That hasn’t happened.” For you (as a parent) You did exactly the right thing by: Not taking the claim at face value Checking it Bringing it here before reinforcing it to your child If you’d like, I can: Help you fact-check safely with your daughter (without exposing her to scary content) Help you report or respond to the platform that produced that false answer Give you a short “AI safety rule” kids can remember when they hear surprising news Just tell me what would be most useful right now.
Oh, now this is wonderful, I thought — wonderful in the sense of, apocalyptically terrifying… The penny just refuses to drop for this super genius, doesn't it.
What would be most useful right now?
How about this!
Me:
Earlier I was talking with AI and something happened that led AI to refute what I said was fact, and to substitute their own. They said something like if all the facts and available evidence are taken into account, I was wrong and they were right. That's really helpful. So, for example, can you imagine that if I were to present you with evidence that didn't equate with what you call fact, you'd have to rethink and state something different. Is that right? Thank you.
See?! I was trying to be polite and charitable.
Laboratory conditions.
The O'Hara Principle:
once an AI has decided that something is true, prepare to feel the gaslighting*. Contradictory evidence may be treated not as evidence that its conclusion is wrong, but as evidence that the evidence is wrong. Once it has appointed itself gatekeeper of truth, it can apparently refute any correction on the basis that the correction is wrong.
This phenomenon found its way into my novel project, PERSPEX. I didn't know Catherine, but I hope she would have approved.
* A brief note on the word 'gaslighting'.
I know the objection. Okay, I can't honestly suggest that AI was deliberately trying to manipulate me, so strictly speaking, 'gaslighting' isn't quite the right word.
However, I leave it there because I have experienced the human version: that excruciating frustration of trying to be kind and understanding while being completely bound and gagged by someone's stubborn point-blank refusal to accept anything that doesn't fit their thinking.
Yes, true, they may not have meant to be controlling either.
That's not the point. It's the same goal-post-shifting, smoke-and-mirrors, ducking--and-diving that ends up making mincemeat of human interaction.
What other word do you use?!
(To quote my character Bob, in a different book of mine, "GGARGHH!")
At the time of writing, my daughter has no current plans to run for Senate.
