Can AI record your meetings without consent? Learn the privacy risks of Fathom, Otter.AI, Google Meet, Microsoft Teams and other AI meeting tools, plus steps to protect your business.
Today, I was in a meeting when a guest I had invited brought Fathom to the meeting, an AI meeting assistant, into our converation.
There had been no discussion beforehand about recording or transcribing the meeting. No request for permission. No explanation of how the information would be processed, stored or potentially shared. This made me think how many businesses are allowing AI meeing assistants into conversations without considering the privacy, confidentiality and potential harm involved for the people who are in that meeting
What concerned me wasn’t that someone wanted to use AI to take meeting notes. It was the assumption that everyone else would be comfortable having their conversation captured and processed by a third-party AI service. Tools such as Fathom, Otter.ai, Loom, Google Meet and Microsoft Teams offers convienient ways to record meetings these days by using artifical intelligence to generate transcripts, summarise conversations, and identify action items.
But, just because these features are available doesn’t mean they should automatically be used for every meeting including job candidates, board room, staff performance appraisals, and the weekly catch up meeting. Are there really any consequences for getting this wrong?
A Real-Life AI Transcript Data Breach: What Happened at a Canadian Hospital?
In September 2024, a hospital in Ontario, Canada, experienced a privacy breach involving an unapproved AI transcript tool. A group of physicians had arranged a virtual meeting to discuss patient cases. One physican who had orginally recieved the meeting invitation had left the hospital more than a year earlier. However, the former physican still had the invitation stored in a personal digital calendar connected to Otter.ai. The AI assistant automatically joined the meeting, transcribed the discussion and generated meeting notes.
Those meeting notes contained personal health information relating to seven patients, including names, diagnoses, medial notes, and treatment information. The AI tool had not been approved by the hospital and the incident triggered a mandatory privacy breach report and an investigation by Ontario’s Information and Privacy Commissioner.
In October 2025, the regulator published its response publicly, highlighting the need for stronger controls over AI transcript tools, meeting access, staff training, and employee offboarding. The hospitial took corrective action, and the regulator recommended addtional safeguards, including a formal request to delete retained patient information.
You can read more about the incident in Field Law’s analysis of the privacy breach and the Ontario Information and Privacy Commissioner’s findings by clicking the link.
Why Should Businesses be Concerned?
This incident did not involve someone deliberatly sending confidential patient records to an unauthorised person, instead, an AI tool connected to a former employee’s personal calendar who was able to access a meeting, process sensitive information and distribute meeting notes.
There are many weakness that contributed to the incident:
An unapproved AI tool connected to a personal calendar
A former employee retaining a meeting invitatin
Insufficient controls over automated meeting participants
Sensitive information being processed by a third-party service
Meeting notes being distributed withouth appropriate organisational oversight
This raises an important question - If your organisation doesn’t know which AI tools are connected to its meetings, calendars, and business applications, how can it know where its information is going?
What Are the Privacy Risks of AI Meeting Recording?
AI meeting assistants are often introduced as productivity tools. To reduce administrative work, allow participants to focus on discussions and produce meeting notes within minutes. However, they can also collect considerably more information than organisation realised.
Depending on the tool and its configuration, this may include audio, video, participant’s identities, meeting chats, screen-shared information, transcripts, summaries, and action items. Consider what might be discussed during an ordinary business meeting:
Customer information and financial circustances
Employee performance, complaints or medical conditions
Confidential audit findings and internal security weaknesses
Commerical negotiations, pricing, and business strategies
Intellectual property and supplier agreements
Legal disputes or regulatory matters
Disagreements in the board room or general meeting room
A meeting recording can become a substantial collection of personal and confidential information without one realising it. And once that information is processed by a third-party AI service, the organisation needs to understand where it goes, who can access it, and what happens to to afterwards.
What Harm Can AI Meeting Transcript Cause?
More than what you think. Privacy risks are not simply about whether an organisation has complied with their legal obligations. They are also about the consequences for individuals and businesses when information is collected, processed or disclosed without appropriate safeguards.
Harm to individuals (those sitting in the meeting)
If confidential conversations are exposed, individuals could experience identity theft, financial fraud, discrimination, embrassment, emotional distress or reputational damage. Image for a second, an AI assistant recording a human resource meeting where an employee’s medical condition, workplace complaint or disciplinary matter is disclosed. If that information is shared wiht an unintended recipient, the consequences coud extend well beyond an administrative error.
Harm to business
Organisations may face financial and reputational consequences if AI transcript tools expose confidential information. This could include loss of customer trust, disclosure of commerically sensitive information, contractual disputes, regulatory investigations, and legal costs. For example, during an audit, an organisation may discuss internal weaknesses, supplier arrangements, business risks or confidential customer information. If an unapproved AI assistant captures that discussion and there was no consent obtained from participants, the organisation may have permitted senstive information to be processed outside its approved systems and controls and without the approval of those sitting in on the meeting (physical or virtual).
Loss of control overy information
Another concern si what happens to the information after the meeting. Does the provider retain the recording? Can the transcript be shared? Is it tranferred internationally? Can it be easily deleted? Is it used to improve or train AI models? The answer depends on the service, its configuration and the contractual agreement in place. Organisations should no assume that all AI meetings assistants handle information in the same way, because they don’t.
Can You Record a Meeting Using AI Without Consent?
This one of the most important questions for organisations using AI meeting transcription. The answer depends on the jurisdiction, the type of conversation, the information being processed and the applicable legal requirements. It is also important to distingish between consent to recording and the lawful processing of personal information.
Australia: Priavcy and AI meeting recording
Under the Australian Privacy Principles, organisations covered by the Privacy Act 1988 (Cth) must consider whether collecting personal information through AI transcript is reasonably necessary for their activities or functions. The Office of the Australian Information Commissioner (OAIC) specifically addresses AI-generated meeting minutes in its guidance on commerically available AI products. The OAIC recommends seeking participants consent as a matter of best practice, particularly because organisations cannot always predict whether sensitive information will be discussed.
Under APP 3, collecting sensitive information generally requires consent unless an exception applies. Australian states and terriotry surveillance and listening device laws may also impose separte requirements for recording private conversations. Organisations should therefore consider both privacy obligations and applicable recording laws before enabling AI transcription. Read the OAIC’s guidance on commerically available AI products.
United Kingdom (UK) and European Union (EU): GDPR Obligations
Under UK GDPR and EU GDPR, organisation must establish an appropriate lawful basis when processing personal data. Consist is one possible lawful basis, but it’s not always required.
Depending on the circumstances, another lawful basis may apply. However, organisations must still meet their obligations relating to transparency, fairness, data minimimisation, security and retention. Where special category personal data is invovled, additional legal conditional may apply. Therefore, organisations should also consider applicable communications recording lawas and contractual confidentiality obligations. Simply activating AI transcription feature does not establish that the processing is lawful.
United States (USA): Recording consent laws vary by state
In the United States, recording laws operate at both federal and state levels. Federal law generally permits recording where at least one participant consents, subject to exceptions. However, some state, including California, Flordia, Pennsylvania, and Washington, impose stricter consent requirements for certain cnoversations. This creates additional considerations for businesses conducting meetings across state boundaries.
For example, an organisation based in Texas might arrange a meeting involving a client in California. The organisation should not assume that complying with Texas recording requirements is sufficient. A practical approach for businesses is seek permission from all participants before recording, while also assessing the legal requirements that apply.
The United States has also seen ligitgation involving AI meeting transcription. In Brewer vs Otter.ai, plaintiffs alleged that conversatins were recorded and processed without proper consent. In August 2026, a federal court allowed certain claims to proceed while dismissing others. This was not a final finding that Otter.ai had violated privacy laws. Nethertheless, the case demonstrates the legal questions emerging around AI meeting recording and the handling of participant information.
How to Use AI Meeting Assistants Safely: Eight Practical Steps
Organisations do not necessarily need to prohibit AI meeting transcription. However, they should establish clear safeguards before allowing employees and contractors to use these tools.
1) Identify and approve AI meeting tools: Maintain an inventory of AI tools used across the organisation. This should include standalone transcription applications adn AI capabilities embedded within existing software.
2) Conduct privacy and due diligence: Before approving an AI meeting transcription tool, establish:
What personal and confidential information it collects.
Where recordings and transcripts are stored and processed.
Whether information is transferred internationally.
Whether the provider uses information for AI training or other purposes.
Who can access recordings and summaries.
What retention and deletion controls are available.
Whether appropriate contractual and security arrangements exist.
Conduct a Privacy Impact Assessment (PIA) or Data Protection Impact Assessment (DPIA) where appropriate or legally required.
3) Inform participants before the meeting: Where AI transcription is planned, explain this in the meeting invitation. Identify the tool, why it will be used, what information will be captured and how that information will be handled. Where consent is legally required, obtain valid consent before recording begins. Where a paricipant objects to the use of AI transcription, it is important to come to an agreement with them before the meeting begins. This may include to stop the recording at certain discussion points, the participants agreeds to depart the meeting for a short period of time and is record that they are leaving the meeting and is again recorded they have returned to the meeting, the particpants agrees not to attend the meeting, or disable the use of AI transcipt during the meeting and note it in the meeting minutes. Even where another lawful basis applies, appropriate transparency remains important.
4) Confirm arrangements before recording starts: The meeting organiser should confirm that participants understand the proposed recording or transcription. For example:
"Before we begin, I'd like to use an AI transcription tool to prepare meeting notes. It will process our discussion and generate a summary. I can explain how the information will be handled, and we can discuss an alternative if anyone has concerns."
Where recording is not appropriate, manually prepared notes may be a suitable alternative. Some AI transcriptions have a pop up notification informing participants if a meeting is held virtually. It is important to provide time and still verbally advise the meeting will be recorded so that particpants can decide weather they consent to the recording or not. Never assume all those that participant automatically consent to AI being used to record the conversation regardless if the discussion is confidential or not.
5) Control automatic meeting access: Review calendar integrations and disable automatic joining where it is unnecessary. Use meeting waiting rooms or lobbies, particularly when sensitive information may be discussed. Meeting hosts should check participant lists and remove unexpected or unapproved AI assistants before proceeding.
6) Restrict access and retentin: Only authorised individuals should have access to recordings, transcripts and summaries. Establish appropriate retention periods, sharing restrictions and deletion procedures. Consider whether retaining a reviewed written summary would be sufficient rather than keeping the complete recording.
7) Train employees and contractors (if applicable): Employees and contractors should understand when AI transcription is permitted, when approval is required and which meetings should not be recorded. Training should cover privacy, confidentiality, supplier obligations and the risks of connecting unapproved AI applications to business systems.
8) Review employees and contractor offboarding: The Canadian hospital incident demonstrates why offboarding must extend beyond disabling work accounts. Organisations should review calendar access, third-party integrations, meeting invitations and delegated permissions. Former employees and contractors should not retain access to confidential business meetings or information.
What Should Businesses Avoid When Using AI Transcripiton Tools?
Some common practices create unnecessary risks.
Do not automatically record every meeting. Recording should serve a defined and justified business purpose.
Do not assume participants are comfortable being recorded. Explain the arrangements before recording or transcription begins.
Do not allow unapproved AI assistants to join meetings. This includes tools connected to personal calendars.
Do not assume existing software approval covers every AI feature. New capabilities may introduce additional privacy and information-sharing risks.
Do not share AI-generated transcripts without reviewing them. They may contain confidential information or inaccuracies. Review the transcripts first before sharing.
Do not retain recordings indefinitely. Establish appropriate retention and deletion requirements.
Do not assume recording consent authorises every subsequent use of the information. Separate processing, sharing or model-training activities may require additional assessment.
Do not treat AI transcription as solely an IT responsibility - It’s not. Business owners, employees, contractors, privacy professionals, legal advisers, information security teams and AI governance functions all have roles to play.
What Should You Do If AI Has Recorded a Meeting Withou Permission?
Discovering that an AI meeting assistant has recorded or transcribed a conversation without appropriate notice, permission or organisational approval should trigger an assessment. However, it does not automatically mean a notifiable personal data breach has occurred. An organisation should take the following steps.
Step 1: Contain the issue. Stop further unauthorised recording or transcription where necessary. Restrict access to affected information and preserve relevant evidence.
Step 2: Establish what happened. Identify which meetings were affected, who participated, what information was captured and where recordings or transcripts were sent.
Step 3: Assess the legal and privacy implications. Determine whether applicable recording laws, privacy requirements, contractual obligations or confidentiality arrangements were breached.
Step 4: Assess potential harm. Consider the sensitivity of the information, who accessed it, whether it was shared further and the possible consequences for affected individuals.
Step 5: Determine whether notification is required. Establish whether the circumstances meet the applicable legal thresholds for reporting a personal data breach or notifying affected individuals.
Step 6: Take corrective action. This may include restricting access, requesting deletion, notifying affected parties, reviewing supplier arrangements and introducing additional controls.
Step 7: Document the assessment. Record the findings, decisions, reasons, corrective actions and responsible owners.
A failure to obtain required recording consent, unlawful processing of personal information and a notifiable personal data breach are distinct legal issues. They may overlap, but one does not automatically establish the others.
Who Is Responsible for AI Meeting Recording and Privacy?
One of the challenges with AI adoption is how easily tools can be introduced without a formal business decision. An employee discovers a useful application, connects it to their calendar and begins using it to record meetings. From their perspective, they may simply be improving productivity. But the organisation may now have another third-party service processing personal information, confidential conversations and business records. This is why AI governance cannot be treated as solely an IT or cybersecurity responsibility.
IT and security teams play an important role in managing technical controls. However, responsible AI adoption requires an understanding of privacy obligations, business processes, supplier arrangements, information use and potential harm. Business owners, Directors, Board Members, and senior leaders need visibility over how AI tools are introduced and used throughout their organisations.
Legal advisers can interpret regulatory and contractual requirements, while privacy and AI governance professionals can support organisations in translating those obligations into practical operational controls. Without clear ownership, businesses may only discover their exposure after something goes wrong.
Frequently Asked Questions About AI Meeting Recording
Is Fathom AI legal to use in meetings?
Fathom can be used lawfully in appropriate circumstances, but legality depends on how the tool is configured and used by the organisation, the participants involved and applicable recording and privacy laws. Organisations should assess their legal obligations, inform participants and establish appropriate safeguards before enabling transcription.
Do you need consent to record a meeting using AI?
Not always. Requirements vary by jurisdiction and circumstances. Some recording laws require consent from all participants, while privacy legislation such as UK GDPR may permit personal data processing under another lawful basis. Organisations must assess both recording requirements and subsequent data processing obligations and document so employees and contractors clearly understand the requirements.
Can AI meeting transcription cause a data breach?
Yes. AI transcription can contribute to a data breach if personal information is accessed, disclosed or otherwise compromised without appropriate authorisation. The 2024 Ontario hospital incident involving Otter.ai is a documented example.
Can employees use AI meeting assistants without company approval?
Not recommended. Organisations should establish clear approval requirements before employees or contractors connect AI assistants to business meetings and calendars. Unapproved tools can introduce privacy, confidentiality, contractual and information security risks.
Final Thoughts: Just Because AI Can Record a Meeting Doesn't Mean It Should
AI meeting assistants can offer genuine benefits. They can reduce administration, improve documentation and allow participants to focus on conversations rather than taking notes. But convenience should not come at the expense of privacy, confidentiality or trust. The Canadian hospital breach demonstrates how an apparently simple AI integration can expose sensitive information when appropriate safeguards are missing.
My own meeting experience raised a more fundamental question:
When did we start assuming that bringing an AI assistant into someone else's meeting was acceptable without discussing it first?
Responsible AI adoption is not about preventing people from using useful technology. It is about ensuring organisations understand what their technology does, how information is processed, what harm could occur and who is accountable for managing those risks.
Before your next meeting, consider three questions:
Do you know whether an AI assistant will be recording or transcribing the conversation?
Do you know where that information will go and who can access it?
Has your organisation established appropriate safeguards for using the tool?
If the answer is no, it may be time to review how AI meeting assistants are being used across your business.
At Hyplon Advisory, we support organisations in establishing practical AI governance and privacy arrangements, including AI tool assessments, supplier due diligence, privacy impact assessments, policies and operational controls. If you have any questions or concerns about the use of AI transcription tools please reach out so we can talk though it.
References and Further Reading
OAIC — Guidance on privacy and the use of commercially available AI products
Information and Privacy Commissioner of Ontario — Hospital privacy breach involving an AI scribe
Field Law — How an AI Transcription Tool Triggered a Privacy Breach
Justia — 50-State Survey of Recording Phone Calls and Conversations
Disclaimer: This article provides general information and does not constitute legal advice. Recording and privacy obligations depend on the jurisdiction, circumstances and applicable legislation.

