Governance and data

What AF9000+ Taught Me About AI Governance

Five habits from aerospace quality auditing: documented work, traceability, risk, regular review and training for practical AI governance.

AF9000+ and AI governance

Quick answer

The useful transfer from quality auditing to AI is a working discipline: define the task and owner, document changes, adapt checks to consequences, review the work and train the people responsible. This first-person reflection offers a practical method, not proof that aviation rules or survey results apply universally.

Want to put these questions into practice with your team?

Explore AI workshops

When people find out I spent twenty-one years in the Canadian Armed Forces doing avionics maintenance, with a few of those years in quality auditing, and that I now work in AI governance consulting, there is usually a pause while they look for the connection. For me, the connection is the operating discipline: a responsibility, a documented process and evidence that the work was checked.

What AF9000+ actually is

The Department of National Defence’s 2022 Technical Airworthiness Manual identifies AF9000 Plus as a quality standard for aerospace engineering and aerospace equipment maintenance. It is a more specific context than calling it the quality system for everything the Canadian Armed Forces does. See section 1.6.2.S1 of that manual, in French.

My experience comes from avionics maintenance, including CF-188 Hornet and electronic warfare work, and from auditing. What stayed with me was the importance of a traceable record behind a sign-off: what was done, against which criteria, what deviation was found and how it was handled. That is the lesson I bring to AI; it is not a claim that an aircraft maintenance procedure can simply be copied into every AI use.

Where the parallels start

The first parallel is documented process. For an AI use, I want a clear description of the task, the information used, the expected result, the person approving the work and the checks that follow. If a result is wrong, that record gives the team something more useful than a memory of what the tool was supposed to do.

The 2025 Gradient Flow/Pacific AI survey illustrates why I distinguish a policy from operating arrangements. It collected responses from 351 participants through newsletters, partners and online advertising between February 14 and May 29, 2025; 91% came from organizations with US operations. Within that sample, 75% reported an AI policy, 54% an incident-response playbook and 59% a governance role or office. These are reported arrangements in that sample, not a measure of all Canadian businesses or the quality of those controls. The survey publishes its method and findings.

The second parallel is the audit trail. Record meaningful changes to the system and its use: what changed, who approved it, when and why. Preserve the relevant test results and decisions. That makes it possible to investigate an unexpected output without relying entirely on the people who happened to be present.

Risk is the third parallel. Review should reflect the consequences of an error and how the tool is used. An assistant answering general product questions and a system supporting hiring decisions call for different questions about information, oversight and escalation. A reassuring product label is not a substitute for examining the workflow.

Regular review is the fourth parallel. I would set review points according to the use, its consequences and changes in the model or data. Check a sample of real work, record problems and confirm whether previous corrections held. A calendar helps make this work deliberate; it does not establish a universal audit frequency for every AI tool.

Training and competence are the fifth parallel. People need to understand the task, the limits of their authority, the information they may use and when to stop or ask for help. Training should include the actual workflow and an opportunity to demonstrate the checks, not only a presentation of the tool’s features.

Keep the discipline in daily work

The mistake I want to avoid is treating governance as a one-time document: write a policy, circulate it for signatures and file the PDF. AF9000+ taught me to look for the connection between a rule and what happens during the work. Who performs the check? Where is the result recorded? What happens when the check fails?

ISO/IEC 42001 offers an AI management framework built around responsibilities, risk management, monitoring and continual improvement. It can support that discipline. It does not transfer aviation requirements into AI, certify the quality of every output or replace the law that applies to a particular use.

I would begin with one workflow, its owner and one example. Write down the expected result, the human check and the path for handling an exception. Revisit that record after use. This is the working method I propose from my experience.

To connect the discipline to everyday information handling, see Loi 25 and AI at work. To clarify what the system may decide on its own, see delegating to AI without constant supervision.

Sources

Sources and scope. Official references are linked beside the relevant passages. Adapt the proposed working methods to your tasks, information and responsibilities.

Start with your team’s work

Explore custom workshops built around relevant situations and appropriate human review.

Explore AI workshops

Back to the articles