PMI's artificial intelligence standard and who answers when it fails

PMI published the first global standard on artificial intelligence in project work. It is not a catalogue of tools nor a guide to prompts. It is the written answer to a question that spent two years without an owner, the question of who answers when the model gets it wrong.

What was published

It is called The Standard for Artificial Intelligence in Portfolio, Program, and Project Management and it came out on 9 June 2026. According to PMI’s own standards page it is the first and only ANSI approved artificial intelligence standard for the project profession, and it brings eight guiding principles and five performance domains, plus a life cycle framework for designing, deploying and overseeing initiatives with AI inside them.

We are not going to reproduce its contents here. The numbers are in the press release and in the standard’s own listing, and that is enough to understand the size of the thing. What is worth telling is why it exists.

Which gap it came to fill

Almost the whole public conversation about AI governance looks at the technology. What a model can do, what it should not do, what ought to be forbidden to it. Very little of it looks at the place where AI actually enters an organisation, which is a project.

Think about it. Every AI system installed in a company arrives through a project. It gets a scope, a budget approval, a team, and someone signs off when it goes into production. The discipline that does that work had no published standard telling it how. There was regulation saying what is allowed and there was the tool doing the executing, but between those two layers sat a gap the size of an entire project.

Pierre Le Manh, who runs PMI, summed it up in a sentence worth keeping. AI transformation succeeds or fails in the projects and programmes that deliver it.

Own diagram based on the standard’s press release and the listing published by PMI.

Why it is technology agnostic

The standard names no models and no vendors, and PMI justifies that without hedging. The guidance has to hold across the AI tools and models that ship in the years ahead.

That decision is less obvious than it looks. A standard written around the capabilities of this year’s models would have aged before it came off the press. By writing it around how the work is approved, governed and delivered, what stays fixed is the procedure and what changes is the tool that goes inside. It is the difference between writing a highway code and writing the manual for one model of car.

Where it fits with the European regulation and with ISO 42001

PMI names both directly. The standard addresses compliance with emerging requirements such as the European artificial intelligence act and the ISO 42001 standard.

Without going into the detail of either, the relationship is quick to grasp. The European regulation sets obligations according to the risk of the system. ISO 42001 describes a management system for the whole organisation. Neither of them tells you what to do on Tuesday in your project status meeting. PMI’s standard steps in exactly there, in the stretch where a general obligation turns into a concrete decision with a name attached.

What changes if you are running a project with AI inside

The main thing is that the standard is built around human in the loop oversight at every stage. Not as a gate at the end, but as a permanent condition of the work.

That is precisely this house’s thesis written in normative paper. The tool executes, the person answers. When a model estimates a duration badly, the responsibility does not dissolve into the model, it stays with whoever accepted the estimate without looking at it. A standard that requires reviewing the output, deciding what gets escalated and deciding when a recommendation is accepted or overridden is saying the same thing in other words.

The practical consequence is more prosaic and more useful. You now have a document to point at when someone upstairs wants to accelerate by skipping the review. That is the real value of a standard, giving you common language to defend a decision in front of legal, audit, finance and technology.

If you are interested in how this lands in the day to day numbers, we worked on it in the post on earned value and decisions with AI.

Join the waitlist

At The PM Architect we write the PMBOK study guides in five languages, with real classroom cases and exercises that come with an explanation. They will be free. Leave your email on the home page and we’ll let you know the moment they’re ready.