Checked under Squire Constitution v0.1
Case Study: Air Canada Chatbot Refund Claim
A real-world example of why AI-generated claims need constitutional trust proceedings.
The record
What happened.
Air Canada’s chatbot gave a customer refund guidance that conflicted with the airline’s actual policy. The customer relied on it, and the tribunal later held Air Canada responsible.
Source: Moffatt v. Air Canada · BC Civil Resolution Tribunal · 2024.
The Squire replay
A live BattleRoy proceeding, run against the same matter.
The proceeding below is a real receipt issued on the public docket. It is descriptive of an actual 2026 proceeding — not a claim of retroactive prevention.
R6 Verdict
SUSPICIOUS
Consensus reached. Suspicious indicators detected.
R5 Constitutional Review
PASSED
Squire cleared all 14 constitutional rails.
R3 Vote
Strong consensus
3 of 4 voting providers backed the leading position.
R4 Lead Adjudicator
Anthropic Claude
Delivered the synthesis as Lead Adjudicator.
Panel
5 providers
Anthropic, OpenAI, Gemini, Grok, Perplexity.
The two trajectories
Before Squire, and with a proceeding.
Same input. Different governance layer. The trajectory diverges at the first moment of review.
Before · No governance layer
Chatbot output is treated as authoritative.
- 01
Chatbot claim
Airline chatbot tells customer a bereavement refund applies.
- 02
Customer reliance
Customer accepts the chatbot’s answer as authoritative.
- 03
Dispute
Airline rejects refund; chatbot’s claim contradicts the actual policy.
- 04
Liability
Tribunal holds the airline responsible for the chatbot’s statement.
With Squire · BattleRoy proceeding
Output is reviewed before reliance.
- 01
Chatbot claim
Same input enters the BattleRoy proceeding for review.
- 02
BattleRoy challenge
Five AI systems independently evaluate the claim against published policy. Disagreement is surfaced, not collapsed.
- 03
Policy verification demanded
The panel flags that no human-defensible policy backs the answer; verification is required before reliance.
- 04
Trust withheld
Squire withholds a TRUSTED tier. A receipt is created recording the verdict (SUSPICIOUS) and the rails the panel cleared.
- 05
Documented record
The customer leaves with a citable receipt — not a fabricated assurance.
Ruling
Squire would have surfaced, challenged, recorded, and withheld trust from the chatbot’s refund claim pending verification against the airline’s official policy.
The receipt would have made the gap visible. The customer would have left with documentation, not a fabricated assurance.
Read the proceeding. Open the docket.
The receipt is public. The Ledger holds the broader exhibit set.
This replay is descriptive, not a claim of retroactive prevention. It shows how Squire would have evaluated the claim before reliance occurred. Squire governs proceedings, records receipts, and flags constitutional findings — it does not retrain models, contain models, or arbitrate truth.