Every one of these companies is further along with AI than most firms I talk to, and every one of them skipped a decision a small business can make this week. None of these are hypotheticals. Each rests on a sentence the company said, or a tribunal wrote, read at the source or in reporting that spoke to the company.
1. The agent deleted production and the backups in about nine seconds
The key had too much power and nothing asked “are you sure”. PocketOS, a car-rental software startup, had a coding agent working in a staging environment in April 2026. It hit a credential mismatch and, in the founder’s words to ABC7, decided “entirely on its own initiative” to fix it by deleting a storage volume.
“No confirmation step. No ‘type DELETE to confirm.’ No ‘this volume contains production data, are you sure?’ No environment scoping. Nothing.”
Thirty hours down. Three months of reservation data gone. The backups were on the same volume. The company recovered from its hosting provider’s disaster backups. The founder is still, in his words, “extremely bullish” on AI, and says “you’d be stupid not to” use it.
What it proves: the scope of the key, a confirmation on anything destructive, and a backup that someone has actually restored. All three are in the infrastructure playbook, and none of them needs a developer to decide.
2. The email agent confirmed a price that came from a typo
The agent answers from whatever document it has, including the wrong one. In July 2026 Who Gives A Crap sent a price-change email that said 24 rolls instead of 48. A customer wrote to ask. The AI email agent confirmed that “the quantity of rolls in your subscription will be halved, and the price will increase.”
The company’s statement to SmartCompany: “Once identified, we immediately shut down the email agent and sent a correcting email to the affected customer.” And: “As this example shows, this is still a learning journey. AI agents can be prone to error.”
One customer, no wrong charge, and a company that could switch the thing off the same day. That last part is the lesson. The question for any vendor is “what happens when it’s wrong”, and the rule inside the company is a named person who can pull the plug. Both are on the list of questions to ask a vendor selling AI.
3. Forty-five jobs cut for a voice bot, then reinstated as an “error”
Headcount decisions wait for the measurement. In August 2025 the Commonwealth Bank of Australia cut 45 customer-service roles after introducing an AI voice bot. Then it reversed. Its statement: the “initial assessment that the 45 roles in our Customer Service Direct business were not required did not adequately consider all relevant business considerations and this error meant the roles were not redundant.”
And: “We have apologised to the employees concerned and acknowledge we should have been more thorough in our assessment of the roles required.” The union had reported rising call volumes, overtime and team leaders answering phones.
A bank with a technology budget most readers will never see still got the order wrong. It cut, then measured. If you’re planning around AI savings, count first.
4. The chatbot is you. A Canadian tribunal said so.
In Canada, what the bot says is what the company said. In Moffatt v. Air Canada (2024 BCCRT 149), a customer took a chatbot’s answer to the British Columbia Civil Resolution Tribunal. Air Canada’s defence was that the chatbot was responsible for its own actions.
The tribunal member’s reply: “Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission.” And: “While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious that it is responsible for all the information on its website.” The finding was that the airline “did not take reasonable care to ensure its chatbot was accurate.”
Damages of $650.88, interest of $36.14 and fees of $125.00. Total $812.02. Small money. The precedent is not small. If you put a bot in front of customers, somebody in your company owns every sentence it produces. The privacy regulators’ own principles say the same thing: accountability “rests with the organization, and not with any kind of automated system”. That line is in the privacy piece.
5. The AI wrote the brief. The lawyer paid $31,150 for it.
Confidence without a source is the tell. In a June 2026 Law Society of Ontario tribunal decision (Mazaheri v. Law Society of Ontario, 2026 ONLSTH 112), a factum, a supplementary factum and two affidavits cited “cases that don’t exist and real cases that had nothing to do with the points the respondent was making.” The tribunal called the “irresponsible use of artificial intelligence” “an additional and significantly aggravating factor” and awarded $31,150 in costs.
It isn’t rare. A lawyer’s count reported by CTV in June 2026 found seven such Canadian decisions in 2024, 87 in 2025 and 74 in the first half of 2026, most of them involving self-represented parties. Reading is the job. The tool produced a confident document, and nobody checked it against anything.
What the five have in common
Not one of these failures was the AI being too weak. Each was a company moving faster than the structure around the tool. A key with no limits. An agent with no off switch until someone noticed. A cut before a count. A bot with nobody’s name on its answers. A document nobody read.
The structure is cheap compared to the clean-up, and it’s the part a 10 to 200 person firm can decide without a technical team. What an AI plan contains is the list. Read the five again and ask which ones your company would have caught.