Three AIs walk into a gym. One shows up with homework nobody assigned. One swears he’s already in shape and pulls something reaching for the first weight. The third — well, you’ll know him when you see him. It’s a joke. It’s also the most honest thing I know how to say about putting AI to work without a way to check it.
What “governance” actually means here
Governance sounds like a 200-page manual nobody reads. In the comic, the trainer reframes it as something you could fit on a whiteboard: four questions you answer before you start, and keep answering as you go. That’s it. Not a compliance department — a habit.
The free starter. Steal it. Tape it above your desk. These are my plain-language version — the same instinct runs through recognized frameworks like the NIST AI Risk Management Framework, which organizes the work into a set of core functions of its own.7
- What are we trying to do?Name the actual goal before you point AI at it. Most bad output is a good answer to a question nobody asked.
- How will we know if it’s working?Decide what “working” looks like up front, so you’re measuring, not vibing.
- What could go wrong?Say the failure out loud while it’s cheap. The airline below is what expensive looks like.
- Who checks the work?A human owns the output. You can’t delegate the checking to the thing you’re checking.
Don’t take my word for it
The comic makes two claims that aren’t jokes. Here’s exactly where each one comes from — quoted word-for-word and linked to the original. Click any of them. Hand the survey to your AI and let it check my numbers, too — though the court ruling sits behind a bot-wall, so that one you’ll want to open yourself. Either way, that’s the whole point: you shouldn’t have to trust me, and you don’t have to.
| What the comic says | What the source says | Source |
|---|---|---|
| 78% of senior leaders lack full confidence their organization could pass an independent AI governance audit within 90 days — and only 1 in 5 have tested a response plan for failure. | “more than three-quarters (78%) lack full confidence that their organization could pass an independent AI governance audit within 90 days”1 · “only one in five has tested a response plan for AI failures”2 | Grant Thornton, 2026 AI Impact Survey (~1,000 senior US leaders)3 |
| In 2022 an Air Canada chatbot invented a refund policy that didn’t exist. In 2024 a tribunal ruled the airline had to honor it. | Tribunal “found Air Canada liable for misinformation given to a consumer through the use of artificial intelligence chatbots”4 after the chatbot gave “incorrect information about the airline’s policy for discounted bereavement fares”5; it awarded “approximately $650 CAD in damages.”6 | Moffatt v. Air Canada, 2024 BCCRT 149 |
Why a comic bothers with citations. Because the joke only lands if the scary part is real. The AI that confidently makes something up and costs its company money isn’t a hypothetical — it’s a decided case. And the four questions aren’t self-help — they’re the cheap version of the audit that 78% of leaders aren’t sure they’d pass. They’re also easy to skip when you’re moving fast. I’ve skipped them.
Nothing on this page is legal or compliance advice; it’s a comic with its homework shown. If you want the real thing for your own situation, the frameworks and the case are linked below — go read them, or point your AI at them.
Sources — quoted verbatim, linked to the original
- 1“more than three-quarters (78%) lack full confidence that their organization could pass an independent AI governance audit within 90 days”Primary
- 2“Nearly three in four organizations are piloting, scaling or running autonomous AI, yet only one in five has tested a response plan for AI failures.”Primary
- 3“based on responses from nearly 1,000 senior business leaders across multiple industries in the US, collected in early 2026”Primary
- 4“found Air Canada liable for misinformation given to a consumer through the use of artificial intelligence chatbots”Reporting the decision
- 5“The AI chatbot provided the consumer with incorrect information about the airline’s policy for discounted bereavement fares.”Reporting the decision
- 6“The Tribunal ultimately awarded Moffat approximately $650 CAD in damages, plus pre-judgement interest and filing fees with the Tribunal.”Reporting the decision
- 7NIST AI Risk Management Framework — the recognized US framework the four questions above rhyme with (further reading, not quoted).Framework