Seven lethal AI failures in M&E — and how to catch them
Nov 10 2026 · Method
Seven AI failure modes in M&E, with the tell for each and the verification fix. Includes a comparison table and pre-sign-off checklist.
The seven lethal AI failures in M&E are: invented figures and sources; false certainty in tone; causation claimed from correlation; missing negative findings; stale or wrong-context data; fabricated citations; and privilege or data-protection breaches. Each has a tell — a number with no source, a confident causal verb, an absence of anything inconvenient — and a fix in the verification routine. Naming the failure modes is the first half of avoiding them; the checklist is the second. The free handbook covers all seven in detail and links to practical templates for checking evidence, logframes and evaluation plans.
What are the seven lethal AI failures in M&E?
AI can speed up monitoring and evaluation work, but it also introduces predictable failure modes. The risk is not that AI is always wrong. The risk is that errors arrive in fluent, confident prose that looks ready to sign off. The seven failures below are the ones most likely to corrupt an evaluation, damage a programme decision or breach a data-protection rule.
The fixes are procedural, not technical. You do not need to become an AI engineer. You need a verification routine that treats every AI output as an untrusted draft until it has been checked against a source, a date and a data-protection rule.
Keep reading — free.
Create an account to unlock the rest of this guide — the worked example, the tables and the checklists — and get The Bridge weekly. Already a member? Sign in.
You'll also get the free handbook by email. Unsubscribe anytime.
Written for M&E professionals. Not a guarantee of accuracy — always verify every claim against its source, and let a named human sign off. Get the free handbook.