A sourced museum of the gap between AI claims and AI outcomes. The captions are sarcastic. The links are not.
Each exhibit links to a regulator, court, company response, or established reporting. We separate documented facts from the joke because hallucinating a scandal would be extremely on-brand and still wrong.
Air Canada’s website chatbot gave incorrect bereavement-fare information. A British Columbia tribunal held the airline responsible for information on its own site.
“Our agent accepted full responsibility. Unfortunately, it has no assets.”OPEN DOCUMENTED SOURCE ↗
Lawyers submitted nonexistent cases generated by ChatGPT in Mata v. Avianca and were sanctioned $5,000.
“Blue links would have reduced innovation velocity.”OPEN DOCUMENTED SOURCE ↗
Google acknowledged odd and erroneous AI Overviews after the 2024 rollout and described changes made in response.
“Optimized for answer presence, not lunch.”OPEN DOCUMENTED SOURCE ↗
McDonald’s ended its IBM automated drive-through test in 2024 while saying voice ordering could still have a future.
“The model upsold you with 99.7% confidence.”OPEN DOCUMENTED SOURCE ↗
The EEOC’s iTutorGroup case alleged recruitment software automatically rejected older applicants; it settled in 2023.
“Bias successfully scaled without increasing headcount.”OPEN DOCUMENTED SOURCE ↗
The IEA projects global data-centre electricity consumption could exceed 945 TWh by 2030 in its base case, with AI an important driver.
“We removed three adjectives and added a power plant.”OPEN DOCUMENTED SOURCE ↗