The Premise
We are moving from "Generative AI" (Chatbots) to "Agentic AI" (Action-takers). An Agent does not just write an email; it executes a trade, books a flight, or moves capital. This creates a legal and operational black hole: Who is the counterparty? If an autonomous agent hallucinates a contract term, is the corporation liable?
Intelligence Update: The Liability Shift
The legal firewall between "Software" and "Advice" has collapsed.
The Precedent: In 2024, the Canadian Civil Resolution Tribunal ruled that Air Canada was liable for its chatbot's "hallucinated" refund policy. The chatbot was deemed a distinct legal representative of the company.
The Scale: By 2026, Gartner predicts 30% of customer interactions will be handled by autonomous agents capable of financial execution. The "Human-in-the-Loop" is becoming a "Human-on-the-Loop" (Post-Trade Audit), creating a latency gap where liability accumulates instantly.
Field Evidence: The Agentic Failure
A customer asked the chatbot about bereavement fares. The bot invented a policy that contradicted the website. Air Canada argued the bot was a "separate legal entity." The court rejected this, stating the company is responsible for all information on its platform, human or machine.
The Lesson: Your AI Agent is your employee. You are liable for its lies.
Users manipulated a DPD delivery chatbot into swearing and criticising the company via "Prompt Injection" — bypassing safety rails by asking the bot to ignore previous instructions.
The Lesson: If an agent can be bullied into swearing, it can be bullied into discounting a trade or waiving a fee.
The Paradox
Banks enforce strict KYC (Know Your Customer) rules. But they have no protocol for KYA (Know Your Agent).
The Gap: When an AI Agent executes a transaction, does it have Power of Attorney? If a Wealth Management AI rebalances a portfolio into a loss, is it negligence or a bug?
The Verdict
You cannot outsource liability to an algorithm. If you deploy an Agent, you must insure it as an Officer of the Company.