A few weeks ago, we did a feature story about whether Canadians should be using AI for health care.

AI is convenient if you don’t have access to a doctor. Plus, it has a pleasant bedside manner.

But Canadians who followed health advice from AI were five times more likely to experience harms than those who did not.

Here’s how it played out in real life according to a couple posts on LinkedIn by Jerrid Grimm.

Following a hockey injury, he went for an MRI on his knee. While waiting for the radiologist’s report, Jerrid took a crack at asking what the different AI models saw.

Steps:
1️⃣ Upload the max allowable MRI images per chat (avg 100)
2️⃣ Ask for an analysis, including severity grading and likelihood of surgery.
3️⃣ Compare results

✴️ Claude Sonnet: Worst-case scenario, everything’s torn, surgery required. 12-16 month recovery.

֎ ChatGPT: Best case scenario, grade 1 sprains, no surgery. Suck it up, you'll be golfing by May.

✦ Gemini 3 and Claude Opus: Somewhere in the middle.

About a week later, the human doctor’s report arrived.

The MRI shows a Grade 2 MCL tear (partial), Grade 1 meniscus (minor), ACL is OK and no surgery recommended. Back to sports after 6-8 weeks of physio.

The winner?

ChatGPT. It still underestimated the MCL and menuscus tears, but at least didn't call for surgery.

Claude Sonnet overestimated everything, calling for certain surgery and almost giving Jerrid a nervous breakdown.

After presenting the results back to the AIs, they all apologized profusely. I'll leave you with this direct quote from Claude Sonnet:

"I catastrophically over-read the ACL, overstated the effusion, fabricated bone contusions at high confidence, and built an entire injury narrative (high-energy valgus mechanism, kissing contusions, hemarthrosis) around findings that the formal radiologist did not confirm. The most alarming aspect is that the errors were reported with high confidence, which is the most dangerous failure mode in any diagnostic context. This analysis would have caused significant unnecessary distress and potentially misguided clinical decisions."

Thanks to Jerrid for letting me share that.

Moving on: Air Canada is back in the news and once again, the subject of our feature story below. Its new complaint process promises faster rulings, but critics say it won’t fix a broken regulator or unfair rules.

-Neil Malik

Air Canada is bypassing a broken system and that could change everything for frustrated passengers. With 95,000 complaints stuck in limbo and waits of up to three years, the airline’s 90-day arbitration test promises speed and transparency the government hasn’t delivered. If it works, it could rebuild trust and reset expectations. If it fails, passengers are back to square one.

Air Canada’s new arbitration plan risks solving the wrong problem and letting Ottawa off the hook. When the airline pays the “independent” mediator, neutrality is immediately in doubt, and flawed federal rules remain untouched. If delays keep getting misclassified, faster decisions may just mean faster denials. The deeper failure isn’t speed — it’s accountability.