AI Governance Looks Good on Paper Until You Ask Who Can Actually Pull the Plug

A new MIT Sloan Management Review piece highlights a critical gap in corporate AI governance: while virtually every large organization claims to govern its AI systems, most leaders cannot name who holds authority to shut down a harmful model. The article argues that the real test of governance is not the existence of policies but the clarity of accountability when something goes wrong.

Why it matters

The MIT Sloan piece cuts through a lot of comfortable noise. Organizations have spent considerable energy writing AI principles, forming ethics committees, and publishing responsible AI commitments. What most have not done is assign clear, tested, operational accountability for the moment a deployed system causes harm. That is a structural problem, not a cultural one, and it will not be fixed by adding another working group.

For those of us in healthcare, this matters more than in most industries. AI tools are moving into clinical workflows, diagnostic support, and patient-facing applications at a pace that outstrips the governance infrastructure meant to oversee them. The question worth asking in your own organization is not whether you have a governance policy. It is whether the person responsible for shutting down a harmful model knows they are that person, has the authority to act, and has ever rehearsed doing it.

What to watch

Whether organizations begin publishing clear escalation and decommissioning protocols alongside their AI governance policies, not just values statements.