Triple A for AI: Assessment, Assurance, Audit
DBTA, Wednesday, November 11th, 2026
AI system development should plan for failure through three linked capacities: assessment, assurance and audit.
Borrowing the American Automobile Association as an analogy, the author says the value of roadside help lies in the assurance that reliable aid is one call away, and AI systems need the same explicit planning for when things go wrong.
That planning comes down to three interconnected capacities, beginning with assessment, the judgment about which assurances a system requires, and assurance, which implements the fail-safes.
Audit is a formal, methodical check that those fail-safes are installed and working, comparable to the test button on a smoke alarm.
Audits are usually tied to regulatory compliance, but enterprises can treat them as integrated health checks, provided they are carried out by parties independent of the system.