Using generative artificial intelligence (AI) technology, automated meeting assistants (aka “AI note takers” or “bots”) are now available to transcribe and summarize online meetings. These assistants may be built into existing collaboration tools such as Zoom, while others are third-party software agents used by individual meeting participants.   

Although these tools have legitimate uses (e.g., for accessibility and capturing notes for later reference), they can also pose substantial privacy, regulatory, and legal risks, and have the potential to stifle conversation and open inquiry. More details on risks are outlined below.  

AI meeting assistants should not be used in Harvard meetings, with the exception of approved tools with contractual protections: 

• Use only AI assistants for which Harvard has an enterprise agreement with the vendor including appropriate security and privacy protections, including:
  • Approved tools as part of limited HUIT-directed pilot programs to evaluate the use of AI assistants within the Harvard environment. You can register your interest using this form.
  • Individuals who wish to request access to AI supported notetaking for accessibility accommodations should contact their Local Accommodations Coordinator to discuss options.
  • To inquire about or request other tools, please contact HUIT. 

We will continue to evaluate their use in our environment. 

• Creating and distributing AI-generated transcripts and summaries could stifle open conversation and discourage some meeting attendees from fully participating, akin to the video or audio recording of a meeting.  
• Personal and Harvard confidential information may be exposed to third parties or used to train AI models, without the necessary contractual protections in place. 
• Discussion summaries may be inaccurate (e.g., “hallucinations” or missing nuances, sarcasm, irony, and important context). 
• Discussion content may be preserved indefinitely or shared with third parties, increasing the risks of inappropriate distribution and unauthorized access. 
• Recorded discussions and summaries/transcripts may be subject to legal discovery in future litigation. 
• Meeting participants may be recorded without their awareness or consent.  

• Use only approved tools with contractual protections: Use only AI assistants for which Harvard has an enterprise agreement with the vendor including appropriate security and privacy protections, including:
  • Approved tools as part of limited HUIT-directed pilot programs to evaluate the use of AI assistants within the Harvard environment. You can register your interest using this form.
  • Individuals who wish to request access to AI supported notetaking for accessibility accommodations should contact their Local Accommodations Coordinator to discuss options.
  • To inquire about or request other tools, please contact HUIT.
• Consider the risks: Before using an AI assistant in a meeting (or when joining a meeting where an AI assistant is active), carefully consider the risks above and ask yourself whether the benefit of using the technology outweighs the risks, taking into account the nature of the meeting. 
• Proactively inform participants: Ensure that, at the beginning of the meeting, all meeting participants are aware of and consent to the meeting being recorded / transcribed. Only begin recording once all participants have consented.   
  • If not automatically announced to all participants, clearly announce your intention to use a bot in the meeting: “Please note that we plan to use an AI assistant to transcribe and summarize this meeting. Please let us know if you have any objections.” 
  • If a participant objects to the use of an AI assistant, do not use the assistant (those receiving the assistant as an accommodation for disability should consult with their Local Accommodations Coordinator about how to proceed in such instances).  
  • When joining a third party’s meeting, explicitly ask if it is being recorded or transcribed.  If the anticipated discussion involves sensitive information, request that the meeting not be recorded/transcribed.  
• Avoid use in sensitive meetings: Explicitly ask participants to disable AI assistants in meetings involving Level 3 or higher information (i.e., sensitive nonpublic information such as patient information, identifiable student information, employee performance or employment actions, personally identifiable records, sensitive research, donor strategy, and privileged legal advice).  This holds true whether the meeting is hosted by Harvard or a third party. 
  • Suggested statement in a meeting when sensitive matters arise: “Given the sensitive nature of this discussion, I suggest that we discontinue the use of AI assistants for the time being.  If there are any accessibility needs, please let us know so we can accommodate them.”  
• Moderate yourself: If an AI assistant is active in the meeting, be mindful of what you say, with the conservative assumption that your words may be preserved indefinitely. 
• Review and correct summaries, share sparingly, and delete when no longer required: Prior to sharing AI-generated meeting summaries, carefully review the document, correcting any errors, then share sparingly on a need-to-know basis.  Keep in mind that summaries may include information that was shared before or after certain participants joined the meeting. Include a notice at the top of the document that the summary was prepared by an AI assistant and may contain inaccuracies, along with instructions for how to submit corrections, e.g. “The meeting notes below were generated by an AI assistant and may contain inaccuracies.  Please submit any corrections to the meeting host.” 
• Delete transcripts and meeting summaries when no longer needed for the purpose for which they were created.  See the General Records Schedule for more information on retention requirements.  
  • If you are subject to a litigation hold, AI-generated transcripts or summaries from meetings you host must be preserved.  
• Be inclusive in your presentation: Review Digital Accessibility Services (DAS) guidance on hosting accessible remote meetings for more information on how to make virtual meetings more inclusive.  

• Prevent AI assistants from joining and expel unwanted assistants: For virtual meetings where you want to ensure AI assistants do not join, enable the waiting room and selectively admit participants (see instructions for Zoom meetings). Third-party AI assistants that are not built into the meeting platform often have an indicative participant name such as “Michael’s OtterPilot”, but this default name can be changed. If an unwanted AI assistant joins, you can remove them from the meeting, or ask that they be removed: 
  • In Zoom, meeting hosts can expel unwanted participants by opening the Participants panel, hovering over the unwanted participant, clicking More, then clicking Remove. 
  • Suggested statement to ask that an unwanted AI assistant is removed from a meeting: “I’m sorry, given the nature of this discussion, we cannot allow AI assistants to record or transcribe this meeting.  If there are any accessibility needs, please let us know so we can accommodate them.”