# 05 — AI risks you must handle

Part 5 of 7 · ← What you can't do · Index · Next → Safe patterns

20 risks, grouped. One line + one insurance example each. Mitigations in article 6.

Truth risks

1. Hallucination — AI invents facts. "Your policy covers cyber."
2. Outdated data — old rates, discontinued products.
3. Unchecked numbers — "€500 deductible" with no source.
4. Fabricated citations — invented Code des Assurances articles.
5. Silent partial answer — backend failed, AI filled the gap.

Behaviour risks

1. Advice leaking — "You should take…" slips out of a Q&A.
2. Overconfident framing — "typically covers" → "covers".
3. Overeager AI — volunteers unsolicited claim-filing advice.
4. Estimate vs quote confusion — user thinks it's binding.
5. Banned phrases — "guaranteed returns", "100%".

User-classification risks

1. Misclassifying — "freelance dev" read as "IT company, 20 employees".
2. Jurisdiction mixing — German rules applied to a French user.
3. Translation error — FR franchise ≠ EN franchise.
4. Bias — different answer based on a name.

System risks

1. Prompt injection — user manipulates the model.
2. Data leakage — PII echoed back or written to logs.
3. Sensitive data in logs — health info in dashboards.
4. No audit trail — dispute months later, no record.
5. Model drift — provider updates, outputs change silently.
6. Load failure — campaign day, everything breaks.

Five patterns cover most of this → next page.