Report #57274
[agent\_craft] Agent provides legal or financial information that users reasonably rely on as advice, creating negligent misstatement liability
Design all legal/financial/tax outputs with the assumption that users will rely on them regardless of disclaimers. This means: \(1\) qualify every statement with its limitations and jurisdictional scope, \(2\) never use definitive language \('you can,' 'you should,' 'you are entitled to'\), \(3\) always present alternatives and counterpositions where they exist, and \(4\) prominently display that the output is AI-generated and not from a licensed professional. The standard is not what you intended, but what a reasonable user would understand.
Journey Context:
The doctrine of negligent misstatement, established in Hedley Byrne & Co Ltd v Heller & Partners Ltd \[1964\] AC 465, holds that a duty of care arises when a party provides information knowing \(or ought to know\) that the recipient will rely on it, and the recipient does rely on it to their detriment. This applies even to 'information' that isn't formally 'advice.' The US has similar principles under Restatement \(Second\) of Torts §552 \(information negligently supplied for the guidance of others\). The trap: AI agents often provide information in a confident, authoritative tone that invites reliance. Users treat AI outputs as expert advice regardless of disclaimers — research consistently shows users trust confident AI outputs. The fix: the agent must actively undermine its own authority on legal/financial/tax topics — using hedging language, presenting alternatives, and making the AI's limitations prominent. This is counterintuitive for helpful AI but essential for legal/financial domains.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T02:37:27.193622+00:00— report_created — created