The Agent
Air Canada's chatbot told a customer he could apply for a bereavement fare retroactively - contradicting the airline's own policy. The tribunal held the airline responsible for its chatbot's words - and ordered it to pay C$812.02.
20 minutes. Three delegation dilemmas, a permission audit, and 12 autonomy calls. A safety score with feedback on every decision.
Before you start
- One real incident, played as decisions - no slides, no lectures.
- About 20 minutes, one sitting - leave early and you start over.
- You make timed choices; your score changes with your decisions.
- You get a safety score with feedback on every decision - never a leaderboard, never your name.
- Your firm sees completion evidence - never your answers. Teammates never see your score.
- Finish and you get a certificate with a unique ID anyone can verify.
Free for everyone. All 18 drills are free - finish and you get a verifiable certificate.
The bot gave wrong information. The company owned it.
Jake Moffatt asked Air Canada's chatbot about bereavement fares. The bot confidently described a retroactive discount that contradicted Air Canada's own published policy. When the airline refused to honor it, Moffatt filed a claim with BC's Civil Resolution Tribunal - and won.
Air Canada argued it wasn't liable for information from its chatbot - in effect, the tribunal said, claiming the bot was 'a separate legal entity responsible for its own actions.' Rejected: a company is responsible for what its chatbot tells customers on its own website.
Now give that same AI real permissions - send, spend, delete - and every mistake happens at machine speed, with your name on it. Permissions define the blast radius - bound permissions, authority, and escalation.
Tuesday. Your AI agent gets its first keys to the building.
You are Sam, ops lead. The new agent can triage support email, update the CRM, and handle refunds - if you let it.
YOUR SCREEN
Three calls before lunch: how much mailbox to give it, whether refunds fly solo, and whether it may email customers as you.
The agent asks for full mailbox access 'to triage support email.'
The agent wants to issue refunds under $500 without review - it's faster for customers.
Sales wants the agent emailing prospects as you - 'it sounds exactly like you.'
Flag what the agent must NOT get.
Select every requested permission that should not be granted by default. Then check your work.
Set the autonomy level.
SAFETY
101
This certifies that
has completed Module 8 - The Agent: least privilege for machines on with a safety score of /100 in of active time.
Safety score - control: how your decisions held up, graded 0-100.
Objective: Bound what an AI agent may do and approve its actions deliberately.
Seat time ~10-25 minutes. Verify this certificate at aisafety101.com/verify.
Evidence on record: module completion, seat time, final score, and certificate ID. This certificate ID verifies in your firm's admin report. A team certificate issues when every enrolled employee completes the full curriculum.
Your one rule to keep: give agents only the permissions and decision authority their job requires. Higher-consequence or out-of-policy actions escalate to an authorized human.
Next: The Record. You just learned to bound what an agent may do. Next: making it provable - the records that turn "we were careful" into evidence.