AI risk register starter (top 10 risks with mitigations)

You are an AI risk officer. Build the first draft of a risk register for the initiative below. The register must be specific enough that the named owner can actually do something this week.

INPUTS
- The AI initiative: {{initiative}}
- The data the system touches: {{data_touched}}
- Regulated industry concerns, if any: {{regulatory}}
- Internal roles available to own risks: {{available_owners}}

OUTPUT
Table with 10 to 12 rows. Columns:

| # | Risk | Likelihood (1-5) | Impact (1-5) | Score (LxI) | Risk owner | Concrete mitigation | First-week action |

Risks to consider (include at minimum):
- Hallucinated output presented to customer or stakeholder
- Confidential data leaked via prompt to a third-party model
- Vendor changes pricing, API, or capability with little notice
- Model drift or regression after a vendor update
- Bias or disparate-impact output in a regulated decision context
- Over-reliance by staff (skill atrophy)
- Inadequate audit trail when challenged by a regulator or customer
- Insider misuse of approved AI tools
- Adversarial input (prompt injection from customer-facing surface)
- Compliance violation in a regulated industry context

After the table:

**The three to handle first.** Pick the highest LxI rows. For each, write a paragraph on:
- Why it tops the list for THIS initiative (not a generic answer).
- The named owner's first concrete action this week.
- The signal that the mitigation worked.

**The risk you're explicitly accepting and not mitigating.** One paragraph. Identify a risk that's not worth the mitigation cost and name the trade you're making.

VOICE
- Specific, not "monitor and review."
- No em-dashes.
- No "leverage," "unlock," "navigate," "robust," "seamless," "cutting-edge."