Nobody in clinical operations is asking whether AI can do the task anymore. They are asking who owns the result when it does. Then they ask it again on the next study. And the one after that.

That was the undercurrent in Clinical Trials Arena’s September 30 coverage of OCT Southern California, where responsible use and ownership sat right next to the capability discussion. It matches what I hear from sponsors and CROs. For a growing list of use cases, from eligibility screening to document review, the technology question is mostly settled. The ownership question keeps getting reopened.

Why AI decisions keep getting reopened in clinical operations

Here’s how it usually goes. A team pilots AI-assisted redaction on one study. It works. People get comfortable. Then a new study starts with a different CRO and a different study lead, and the whole conversation starts over. Who reviewed the output? Who approved using it? Does that approval carry over, or does someone need to sign off again?

Nobody can answer quickly, because the original decision lived in an email thread and a meeting nobody wrote down. So the team does the safe thing and checks everything by hand. The AI is still running. It just isn’t saving anyone time.

That’s the cost nobody puts on a slide. It isn’t a dramatic failure. It’s a slow tax on every study, paid in rework and second-guessing.

AI labs are writing their own rules. Trial teams are rewriting theirs every study.

On September 29, the largest AI labs signed a voluntary safety accord at the White House. Google, OpenAI and Anthropic are also reportedly moving ahead with their own industry body to set safety standards for powerful models. You can debate whether voluntary is enough for them. What struck me was the contrast. The labs are at least trying to decide things once. Most trial organizations are making the same AI decisions over and over, one study at a time, team by team.

Where AI ownership breaks down: the handoffs between sponsor, CRO and site

The problem usually isn’t inside any single organization. It’s at the handoffs. A sponsor approves an AI step. The CRO runs it. The site sees the output. Each one holds a piece of the decision and nobody holds the whole thing. When a question comes up, people reconstruct the answer from memory and inboxes.

How a governed workflow lets AI decisions stick

This is the layer we have been building Judi around. Process, people and compliance designed together. When an AI step sits inside a governed workflow, the decision about how it’s used gets made once and travels with the process: who owns it, which version was approved and what happened at every handoff. The next study doesn’t start from zero. It starts from what you already decided.

Outside a governed workflow, every AI step is a decision someone will have to make again.

Why governed AI makes trial teams faster, not slower

Governance has a reputation for slowing things down. In my experience it does the opposite. The teams I see moving fastest with AI aren’t the ones running the newest tools. They’re the ones who stopped relitigating. They decided who owns what, built it into how the work actually runs and moved on.

Everyone else is still having the same meeting.