When AI Decides, Who Is Accountable?

I think it is a design flaw to have AI make decisions with no human oversight. If we take the product prioritization as an example, the Product Owner will still have responsibility for managing the backlog and priority. The Product manager might use AI as a tool to support their work, but it should never take over the work entirely.

Good question, Shenila . Even when AI makes decisions, accountability still rests with people - particularly the project leaders and the organization that chose to implement the tool. AI can support or automate decisions, but it cannot own the outcomes. If decisions lack traceability or clear ownership, that points to a governance gap, not an AI issue, and it’s up to leadership to ensure transparency and accountability remain intact.

I recall the case of a Nascar race in which AI suggested a strategy to win the race. The team did not agree with the views of AI and took another approach. Long story short, they lost the race and realized than would have they followed what AI suggested, they would have won. This is to say that humans can´t compete at processing data with machines and AI. However, the accountability of the decision made (ignoring AI, but also if they would have decided otherwise) remain entirely with us, the humans.

When AI-driven systems begin influencing project decisions without explicit human authorization, the ethical responsibility does not disappear but becomes diffused.
As Shenila insightfully highlights, modern delivery environments increasingly rely on algorithmic engines to reorder backlogs, reassign resources, adjust timelines, and flag risks. These tools operate with remarkable efficiency, yet their opacity can blur the line between human judgment and machine-generated outcomes.
In such situations, the ethical burden ultimately rests with the humans who select, configure, oversee, and rely on these tools. AI may automate the mechanics of decision-making, but it cannot carry moral agency.
Ethical responsibility remains with project leaders, sponsors, and governance bodies who must ensure transparency, traceability, and accountable oversight. This is not merely an academic concern; it is a daily professional obligation.
As practitioners, we must master ethics, not treat it as a periodic compliance exercise. Ethical behavior is a discipline we apply in every meeting, every decision, every interaction. Like you, as a committed member of the PMI Ethics Advisory Team, I see firsthand how essential it is to maintain clarity of authorship, uphold integrity, and safeguard trust when emerging technologies reshape our work.
AI can support decisions, but it must never obscure accountability.
My call to all project professionals:
-- stay vigilant
-- stay curious
-- stay ethically grounded
Be mindful of both real and perceived ethical dilemmas.
Question the invisible logic behind AI recommendations.
And above all, ensure that human responsibility, not algorithmic convenience, remains at the center of our practice.

Shenila, you're naming something real, but I think there are two distinct problems tangled together here, and they have different answers.

The first is accountability. Who is legally and operationally responsible for the outcome of a decision. This one has a clearer answer than the rest of your post implies. AI cannot be accountable because it cannot bear consequences. Accountability rests with whoever deployed the system, set its parameters, and approved its use in the workflow. If an organization has not made that ownership explicit before the AI starts making calls, that is a governance failure, not an AI problem. The diffusion of accountability you describe is real, but the cause is human (someone failed to define ownership), not algorithmic.

The second is traceability, or what you called "ethical authorship." Can you reconstruct why a decision was made, after the fact, in enough detail to audit it. This is the genuinely open problem, and it is where PM leaders should be focused.

Separating these makes the work concrete:

• For accountability, name the human owner before the AI ships, not after something goes wrong. RACI does not hold up when one of the parties is software. Newer frames like Owner / Vetoer / Implementer / Support exist for exactly this reason.
• For traceability, insist that every AI-generated decision leaves a log a human can read in plain language, with the inputs it considered and the alternatives it rejected. If the system cannot produce that on demand, it does not belong in production.

The hardest part is not technical. It is the discipline to define ownership and audit standards before deployment instead of negotiating them after a bad outcome.

Accountability still belongs to project leaders and organizations, even when AI supports decision-making. For example, if an AI scheduling tool reallocates resources and causes a project to fail, the project manager and governance team remain responsible because AI provides recommendations, but humans approve and oversee the final outcomes.