Recent Resource
Guides
When AI Goes Wrong: Turning AI Harm into an Incident Response Process
This three-part white paper series examines AI harms through an operational incident-response lens.
Each paper provides expert insights on how organizations define AI incidents, build a structured and defensible response framework, and make critical decisions
about causality, severity, escalation, and reporting.
Part I: Turning AI Harms into Operational Risk
Responsible AI principles are essential, but principles alone cannot guide an organization through a real-world incident.
As AI becomes more deeply embedded in business operations, organizations need a consistent way to recognize, classify, investigate, and respond to AI-related harms, hazards, and near misses.
This white paper explains why AI harm should be treated as an operational incident response challenge and how organizations can begin building a shared foundation for action.
KEY TAKEAWAYS
• Understand why AI harm is an operational risk, not solely an ethics concern
• Learn what should qualify as an AI incident, hazard, or near miss
• Explore how AI incidents differ from traditional privacy and security events
• Establish a common internal vocabulary for investigation, escalation, and reporting
Part II: Creating a Defensible AI Incident Response Framework
When an AI-related concern emerges, inconsistent intake and ad hoc escalation can leave evidence fragmented, responsibilities unclear, and broader risk patterns undetected.
This white paper provides a practical framework for capturing, investigating, escalating, and remediating AI incidents.
Learn how to connect AI governance with legal, privacy, security, compliance, product, and technical teams through a repeatable, auditable response process.
KEY TAKEAWAYS
• Identify the minimum information every AI incident intake should capture
• Decide whether your program will investigate harms only or include hazards and near misses
• Translate organizational harm standards into reusable assessment criteria
• Create severity levels that trigger consistent ownership, investigation, and escalation
Part III: Special Considerations—Causality and Severity
Did an AI system cause the harm, materially contribute to it, amplify it, or fail to prevent it?
And how serious was the actual or potential impact?
Causality and severity are two of the most consequential and difficult questions in AI incident response.
This white paper offers practical guidance for evaluating those questions, preserving relevant evidence, and documenting decisions that can withstand regulatory, legal, and organizational scrutiny.
KEY TAKEAWAYS
• Evaluate direct causation, material contribution, amplification, and failure to prevent harm
• Assess severity using real-world impact, scale, duration, and reversibility
• Understand how litigation and regulatory reviews may approach AI incidents differently
• Build evidence readiness with consistent, time-stamped, and auditable records
Get Your Personalized Walkthrough
