At a Glance
- The increasing popularity of AI Meeting Assistants makes it essential for companies to create and implement a written corporate policy that establishes the “dos and don’ts” for AI Meeting Assistants, taking into account individualized business operations, industry standards and legal/regulatory requirements.
- While the power of a new AI Meeting Assistant is enticing, the same considerations that led meeting participants to refrain from using audio or video recording should apply, and potential practical and legal considerations must be considered when deciding whether to invite an AI Meeting Assistant to the discussion.
- Lastly, organizations should take the time to review and negotiate adequate terms: (a) with AI Meeting Assistant vendors, to mitigate the risk of confidential, privileged or otherwise proprietary information being released or used to train AI technologies; and (b) with your other service providers to mitigate the risk of third-party initiated recordings that do not comply with your policy.
A “low rumple,” followed by a “metallic ‘squink,’” a “mysterious ‘glonk,’” and someone yelling “Dear God!”
After over-thinking an ill-timed throat-clearing by a board member of the Susan Ross Foundation, a paranoid and insecure George Constanza plants a hidden tape recorder at the next board meeting to determine how they really feel about him. George, not surprisingly, ends up with a fairly useless tape, a torn briefcase, and a failed scheme with no answers (but a hilarious story).
At first glance, recording, transcribing and using tools to summarize meetings appears to be a no-brainer for operational efficiency (and in George’s case, an easy way to spy on the board members). Now, modern artificial intelligence applications (AI Meeting Assistants) have made these functionalities even more accessible. However, the addition of AI Meeting Assistants to traditional all-in-one communication platforms (e.g., Zoom, Google Meet, Microsoft Teams), raises various legal and business questions that require careful consideration.
As discussed in this article, the increasing popularity of AI Meeting Assistants makes it essential for organizations to create and implement a written corporate policy that establishes the “dos and don’ts” for AI Meeting Assistants, taking into account individualized business operations, industry standards and legal/regulatory requirements.
Understanding the Technologies’ Capabilities
Traditionally, written minutes taken during meetings consisted of the date, time, title, attendees, and a summary or recap of each speaker’s input. Meeting minutes provided an account of the proceedings, with a sometimes comprehensive and sometimes sparse sense of what was said by the attendees and what was decided. However, this method often fell short in accurately capturing the specific ideas and nuances conveyed by the speakers, as meeting minutes were generally drafted by an attendee — participating and notetaking at the same time. While the technology for taking an audio or video recording of a meeting has existed for decades, in most circumstances meeting participants opted out of using recording devices, instead favoring pen and paper, often because reviewing recordings takes too long.
The reluctance to capture the entire audio or video transcript may be waning due to enhancements offered by AI Meeting Assistants that go far beyond crude recording technology. There are now a proliferation of AI Meeting Assistants available that can “listen” to a meeting and prepare real-time transcription, tailored summaries and action-item extraction; send post-meeting recap emails; and even analyze the emotional tone of a discussion or the level of engagement of meeting participants.
A “recording” of a meeting refers to capturing the entire audio and/or video of a meeting. This method is particularly useful for noting tone of voice and nonverbal cues, as well as in legal cases where a full recording is required. “Transcribing” a meeting involves converting the recording of the entire meeting into written text, which allows for easier capture of key points and dialogue, as well as improved searchability and review. Lastly, the concept of “note summarization” takes transcribed texts and generates a concise summary by identifying the important points and details, effectively distilling the information into a shorter version. AI Meeting Assistants can do all the above.
While the power of a new AI Meeting Assistant is enticing, the same considerations that led meeting participants to refrain from using audio or video recording should apply, and potential practical and legal considerations must be considered when deciding whether to invite an AI Meeting Assistant to the discussion.
Considerations for Using AI Meeting Assistants
When determining which meetings to record, transcribe or summarize, organizations should consider several factors, including the following.
Transparency and Consent
In the United States, at least one party must consent to recording a conversation.1 In some states the consent of all meeting participants is required. These consent obligations may also apply to the operation and output of certain AI Meeting Assistants. Therefore, it is crucial for organizations to understand how their respective jurisdictions approach consent and to ensure proper consent is obtained when using AI Meeting Assistants. To maintain transparency and trust, when circumstances permit the use of AI Meeting Assistants, the “host” should inform all meeting participants that an AI Meeting Assistant is being used, what outputs it will prepare, and any potential limitations it may have.
Accuracy and Liability
Inaccurate or incomplete output produced by AI Meeting Assistants can lead to misunderstandings or misrepresentation, especially if the text is later relied upon for decision-making or compliance purposes. To avoid liability or embarrassment from relying on misleading excerpts, AI Meeting Assistant-generated output should be reviewed immediately after generation. If any incomplete or inaccurate statements are identified, careful consideration should be given to whether it is appropriate to correct and document the changes, as altering these records may present other issues. In any case, the output should be accompanied by appropriate disclaimers indicating that the output was AI-generated and may not be comprehensive or accurate.
Bias and Discrimination
AI Meeting Assistants may unintentionally introduce bias in how meeting discussions are captured. The technology may struggle to recognize or accurately attribute speech for certain speakers with accents, non-native English vernacular, or unique pitches or cadence. This can result in an unfair disadvantage to certain participants, potentially leading to discrimination claims. When using an AI Meeting Assistant during a meeting with several participants, organizations should consider what strategies can reduce the risk of bias being introduced in the meeting transcript or summary.
Privacy and Data Security
When AI Meeting Assistants join a call, record it or later produce a corresponding summary, there is a risk of access by unauthorized parties. Also, third-party vendors are often involved in recording calls, storing the recordings or producing texts. Organizations should be aware of the access controls their vendors employ and understand how these parties handle cyber threats. When using a third-party vendor, ensure strict contractual limits on the vendor’s access to data, with clear rules on data use and protection in case of a breach. Organizations should also evaluate the necessity and risks of memorializing discussions involving proprietary, confidential or trade secret information.
Training the AI Language Model
Sometimes the vendors of AI Meeting Assistants will seek to use records and generated texts to train their language model. This could inadvertently make confidential, privileged, or otherwise sensitive information public or available to other customers. Therefore, organizations should ensure (by way of due diligence and contractual language) that the vendor they employed does not use meeting information to train the overall language model.
Data Retention and Deletion
Sometimes third-party vendors will store the recordings used to produce the AI-generated transcript or summary. Organizations should evaluate the current requirements of their document retention policies, and, if necessary, establish new policies regarding how long the recordings and AI-generated transcripts are retained and when they are securely deleted (particularly for sensitive or privileged information). Note that regulatory requirements may also drive these retention policies.2
Future Litigation
Assume all AI recordings, notes and transcriptions are discoverable in litigation. These materials could serve as helpful or harmful evidence, depending on what is captured. For example, notes from a project implementation call could demonstrate a party was warned of certain risks or failed to take suggested actions but could also show the organization was silent on a key issue or concern. In any case, AI-generated texts related to matters in litigation must be preserved to comply with legal holds and discovery obligations. Failure to do so can result in sanctions or claims of spoliation of evidence. The systems that store the notes must be capable of overriding any automated deletion schedule to comply with legal holds.
Attorney-Client Privilege
Stenographers in most instances are considered ministerial agents of attorneys, meaning that communications between a client and their lawyer in the presence of a stenographer are typically protected by attorney-client privilege, as the stenographer is necessary to facilitate the legal representation by recording the conversation and maintaining confidentiality.3 Whether such protections will extend to technologies that function in a similar capacity is unclear. Therefore, organizations may consider disabling recording and transcription tools during privileged conversations to avoid discoverability or inadvertent waiver of privilege. Any notes of privileged conversations should be safely stored and labeled as attorney-client privilege. Additionally, attorney notes can be subject to work product privilege, and the work product of an algorithm could be subject to privilege challenges.
Is a Written Record Desirable or Advisable?
Prior to the existence of AI Meeting Assistants, there were circumstances where due to the sensitivity of subject matter, organizations would decide not to prepare a written record of a meeting or to impose limits on who would be taking notes at a meeting. The ease of use for AI Meeting Assistants can lead to default adoption and overuse, which may not always be productive for candid conversation. Organizations should therefore “read the room” and consider the costs of participants’ apprehension before using these tools in meetings.
Conclusion
Organizations should consider developing policies that set forth the appropriate use of AI Meeting Assistants, which may include:
- Safe harbors that allow for the use of the tools in approved circumstances
- Outlining circumstances for when recording is permissible and when prior approval is required (and by whom)
- Prohibitions on use in higher-risk circumstances, such as privileged conversations
- A continuous update and review process to ensure alignment with legal and regulatory considerations
When determining whether recording is permissible or not, organizations should also consider the scope of the meeting, the identity of the attendees, the subject matter, and other regulatory and compliance obligations. Different industries and subject matter may require different approaches.
Lastly, organizations should take the time to review and negotiate adequate terms: (a) with AI Meeting Assistant vendors, to mitigate the risk of confidential, privileged or otherwise proprietary information being released or used to train AI technologies; and (b) with your other service providers to mitigate the risk of third-party initiated recordings that do not comply with your policy.
For More Information
Companies utilizing AI — whether internally, externally or in coordination with their vendors — should monitor developments in this space.
