What happened
The Information and Privacy Commissioner of Ontario reported in October 2025 that Otter.ai, an AI transcription tool, joined and recorded an online hepatology rounds meeting at an Ontario hospital on 23 September 2024.
According to the commissioner’s letter, a doctor had left the hospital in June 2023, but his personal email address was still on the meeting invitation. The letter records that he installed Otter.ai on a personal device in September 2024. When the hospital meeting took place, the tool joined using the meeting link saved in his personal calendar.
After the meeting, Otter.ai automatically emailed a meeting summary and access to a transcript, addressed to 65 users on the recipient list. Staff had discussed personal health information about seven patients during the meeting.
The letter records that some recipients confirmed they had deleted the information or had not received it. Others appeared to have left the hospital without confirming whether they had deleted it. The former doctor did not respond when asked to contact Otter.ai to request deletion of the information.
The hospital had not approved Otter.ai for use in its meetings.
What an auditable version would have shown
The commissioner’s letter reported what the hospital was able to establish after the incident, including who was sent the email, who confirmed deletion or non-receipt, and which patients had been discussed. The recording came to light when Otter.ai sent its email after the meeting.
In this entry’s reading, the hospital needed a record of who had access to the meeting and what happened while it was taking place, rather than having to piece everything together afterwards.
A record made at the time would have shown who was on the invitation, whether they still worked at the hospital and who was admitted to the meeting. It would also have identified any AI tools that joined, whose accounts they were using, what they recorded and where they sent the information.
The hospital could then have used those records to establish who had received the patient information and whether it had been deleted.
The sources reviewed do not report whether the meeting platform logged that a transcription tool had joined, or whether anyone in the meeting noticed it before the email arrived.
Where the gap was
The letter reported that Otter.ai joined through the meeting link in the former doctor’s personal calendar and then sent information through his account.
In this entry’s reading, the doctor had left the hospital more than a year earlier, but he was still on the invitation to a meeting where confidential patient information was discussed. The tool he installed was able to use that invitation to join the meeting.
The published record does not identify an admission control that stopped an automated participant joining on his behalf.
An EgressGate is designed to check whether information contains sensitive data and whether it is being sent outside the trust boundary. In this case, Otter.ai was running through the former doctor’s personal account. The hospital needed to control who could enter the meeting before patient information could be recorded.
A ConductRecord keeps a record of each action an AI tool takes, including whose account it used, what information it captured and where it sent that information.
EgressGate and ConductRecord are Headlights designs.
What governance should have looked like
The commissioner reportedly recommended that the hospital contact Otter.ai directly to request deletion of the information. The recommendations also included reviewing how access is removed when staff and doctors leave, enforcing meeting lobbies whenever health information is discussed, clarifying the rules on personal devices and aligning the hospital’s AI governance framework with the commissioner’s presentation on AI in Ontario’s health sector.
Where clinical meetings are held online, best practice would be to remove former staff from meeting invitations when they leave the organisation. Every participant, whether a person or an AI tool, should be checked before being allowed into a meeting, with a record of who admitted them.
Failure Pattern: a physician who had left the hospital more than a year earlier remained on a clinical meeting invitation, and an AI transcription tool he later installed joined through it and sent patient information to the recipient list.
Governance Principle: an organisation that discusses health information in online meetings should be able to show who was admitted to each meeting, human or automated, on whose authority, and that people who have left can no longer be admitted through invitations they kept.
The reference implementation of EgressGate and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.
Sources
- IPC response to hospital privacy breach involving an AI scribes tool, file HR24-00691 (Information and Privacy Commissioner of Ontario, 27 October 2025)
- Letter to the health information custodian, file HR24-00691 (Information and Privacy Commissioner of Ontario, 27 October 2025)