Three Frameworks Help CISOs Manage Liability From AI Deployments

Summary
Cyber legal expert Daniel B. Garrie outlined frameworks for scoring AI deployment risk, documenting board decisions, and preserving evidence during incidents, alongside contract provisions and exercises for CISOs.
Key points
- Garrie advises negotiating AI vendor contract terms early, including access to logs, model snapshots, preservation requests, and clear risk allocation.
- His five-factor risk rubric assesses potential harm, data and rights, autonomy, evidence gaps, and vendor dependence to guide approval decisions.
- A board-decision record should capture the deployment scope, risks and evidence, approval conditions, accountable parties, challenges, and oversight.
- The incident-response checklist covers AI system state, interaction records, data and access information, governance, integrity, and chain of custody.
- Garrie recommends testing the frameworks on an AI deployment, improving a board report, and running an AI tabletop exercise within 30 days.
Article Details
- Defense Focus
- Assess AI deployment risk and preserve defensible evidence before and during an AI-related incident.
- Detection Methods
- Assess deployments across potential harm, data and rights, autonomy, evidence gap, and vendor dependence.
- Use an AI incident-response checklist to preserve system state, interaction records, and related evidence.
- Run an AI tabletop exercise to practice coordinating vendor evidence preservation and notifications.
- Data Sources
- Security logs
- AI system state at the time of the incident, including a model snapshot
- AI interaction records
- Data and access information
- Governance records
- Board decision records, including risk evidence, approval conditions, oversight, challenges, and accountable parties
- Data integrity and chain-of-custody details
- Defensive Actions
- Negotiate AI vendor contract provisions for access to logs, preservation requests, and timely snapshots of the model at the time of an incident.
- Clarify vendor AI-use permissions, data restrictions, warranties, insurance, liability caps, and operational security responsibilities before signing.
- Record board decisions with their scope, time frame, supporting risk evidence, deployment conditions, human-approval requirements, oversight, and due dates.
- Preserve the AI system's incident-time state and associated interaction, data, access, governance, integrity, and chain-of-custody evidence.
- Involve legal counsel immediately during an AI incident.
- Test the frameworks on an AI deployment, improve a board report, and conduct an AI incident tabletop.
