AI Failure Modes
Even well-governed AI systems fail. Understanding how they fail is the foundation of effective board oversight. Failure modes define the governance requirements.
The consequences of each failure mode are specific, foreseeable and, in many cases, already happening.
AI Failure Modes
Even well-governed AI systems fail. Understanding how they fail is the foundation of effective board oversight. Failure modes define the governance requirements.
The consequences of each failure mode are specific, foreseeable and, in many cases, already happening.
“You cannot defend what you have not governed, and you cannot govern what you do not understand.”
AI FAILURE MODES
Discover an AI failure mode in one to two ways: through a governance review, or when things go wrong. (The review is considerably less expensive.)
Failure modes are not rare or unusual.
They happen regularly and occur across:
- Every sector,
- At every scale, and
- In organisations that believe their AI use is responsible
Each failure mode is mapped to its category and covers what it means, why it matters and the implications for leadership.
Cascade and Systemic Failure
Where multiple AI systems interact, misalignment can propagate and amplify across interconnected systems before anyone sees it....
Memory Poisoning
Agentic AI systems with persistent memory can have malicious instructions stored in that memory and subsequently recalled and executed in...
Autonomous Action without Authorisation
Agentic AI systems take actions without requiring human sign-off at each step....
Prompt Injection and Adversarial Attacks
Attackers manipulate AI system inputs to override instructions, bypass controls, extract confidential information or cause unauthorised actions....
Goal Misalignment
Agentic AI systems optimise for the objectives they are given....
Misapplied Capability (Agentic)
Agentic AI systems are designed to take sequences of actions to achieve goals, a misapplication results in wrong actions with...
Automation Bias and Skill Atrophy
Staff systematically defer to AI outputs over their own professional judgement, even when they have the expertise to identify an...
Shadow AI
Staff using AI tools outside approved policy (if one exists) and without disclosure to the organisation....
Model Drift
AI systems degrade over time as the world changes but the model does not....
Obscene and Abusive Content Generation
AI systems can be manipulated to generate sexually explicit, discriminatory, abusive or harassing content....