PulseAI models and training data
Whether your data is used to train models, how to change the setting, and the position under copyright law.
Whether the meeting audio, transcripts, notes, uploaded material and chat exchanges held in Pulse are used to train AI models depends on your plan. This article covers how each plan is treated, what the setting changes, where to change it, and how it is analysed legally.
Team plan Enterprise plan
Data held on business plans (Team and Enterprise) is not used to train models. On the free and Personal plans it is on by default, and users can switch it off at any time. How each plan is treated is set out under “How each plan is treated”.
What the setting changes
The table below describes the free and Personal plans, where the setting can be changed. Business plans are treated the same as the “off” column.
| Setting | How data is handled |
|---|---|
| Off | data is never used to train models. It is used only for the processing that provides Pulse's features — transcription, generating summaries, generating chat answers, search and so on |
| On | data may be used for training that improves Pulse's accuracy. Even then, it is processed so that individuals cannot be identified |
Under either setting, data is never used to train an external model provider's models. UNCHAIN, Inc. has it in its contracts with the providers it engages for transcription and summary generation that data sent to them is not used in their own machine learning.
Statements about use for training appear in three places: this article, Security and privacy FAQs and the Privacy policy. Check UNCHAIN, Inc.'s implementation and keep all three consistent. Two things in particular need confirming.
- That the treatment per plan (on by default for free and Personal; not used on business plans) matches the implementation. Align the plan names with the actual plans too.
- Whether “processed so that individuals cannot be identified” amounts, in practice, to pseudonymously processed information under Article 41 of the Act on the Protection of Personal Information, to anonymously processed information under Article 43, or to no more than stripping identifiers without reaching either. Anonymously processed information has processing standards set by the Enforcement Regulations, and free-form meeting speech does not meet them easily. If you cannot confirm it, avoid the word “anonymised” and write “processed so that individuals cannot be identified”, as the table above does.
How each plan is treated
| Plan | Use for training | Who can change it |
|---|---|---|
| Free and Personal plans | on by default | the user can switch it off at any time under Settings → Preferences → Data and sharing |
| Team plan | not used for training | this would go beyond the scope of the delegation, so it is off by default and there is no setting for users to enable it. Administrators can check the workspace-wide setting |
| Enterprise plan | not used for training | as above. This would go beyond the scope of the delegation, so it is off by default and there is no setting for users to enable it |
For data held on business plans (Team and Enterprise), UNCHAIN, Inc. acts as a party to whom the customer has delegated the handling of personal data. A processor cannot handle personal data beyond the scope of the work delegated to it, so within that framework use for training is not an option. The basis is explained under “Data held under delegation”.
Changing the setting
The setting can be changed on the free and Personal plans. On business plans the option does not exist at all, so the steps for an administrator to check the current state are given too.
Switching it off on the free and Personal plans
Open settings
Open the Pulse desktop app and open Settings.
Choose Preferences
Choose Preferences from the menu on the left.
Find the data and sharing item
Find the setting for using data to improve models, under Data and sharing.
Switch it off
Switch the toggle off. The change takes effect immediately, and notes and transcripts created afterwards are not used for training.
- the model improvement toggle under Data and sharing is off
- you switched it off before recording a meeting involving anything sensitive
Administrators checking the workspace-wide setting
On business plans, use for training is off by default and there is no setting to enable it. Administrators can check the workspace's current state as follows.
Open workspace settings
From Settings, open Workspace, then General.
Open the data security item
Check the state of use for model training under Data security.
Use what is displayed for internal purposes
What is displayed can be used as-is in answering internal security reviews and processor assessments.
The toggle for using data to improve models, with explanatory text beneath it.
Changing the setting is not retrospective
Switching the setting off on the free or Personal plan means notes and transcripts created afterwards are not used for training. UNCHAIN, Inc. cannot, however, retrospectively determine on an individual basis whether data from before the change was used for training. On business plans the question does not arise, because data is not used for training at all.
Data already used in training cannot technically be removed from a model. If you plan to use Pulse for a meeting involving anything sensitive, switch the setting off before recording that meeting.
The models we use
Pulse uses a combination of models to generate summary notes. Different models suit different kinds of meeting and purposes, so UNCHAIN, Inc. evaluates them continuously and switches between them.
When asking about meeting content in chat, you can choose the model. For the list of models available and what each is like, see Choosing a model.
At present you cannot connect your own model to Pulse. Transcription and summarisation use models selected by UNCHAIN, Inc.. We are considering it as a feature for business customers, but there is no timetable.
The position under the Act on the Protection of Personal Information
Under the Act, using meeting data for training raises two questions: whether it falls within the purpose of use, and whether it falls within the scope of the delegation. We take them in turn.
Relationship to the purpose of use
Meeting audio, transcripts, notes and speaker names can readily be cross-referenced with attendees' names and calendar information, so they normally constitute personal information.
When handling personal information, the purpose of use must be specified as far as possible, and handling it beyond the scope necessary to achieve that purpose requires the individual's consent as a rule. Where meeting data is used to train AI models, wording such as “to improve our service” is likely to be insufficiently specific, so UNCHAIN, Inc. treats use for training as a separate purpose of use and writes out the categories of data covered and whether the setting can be changed.
The purposes of use are set out in full in our Privacy policy.
Data held under delegation
Provision of personal data accompanying delegation within the scope necessary to achieve the purpose of use does not constitute provision to a third party. A processor, meanwhile, cannot handle that personal data beyond the scope of the work delegated to it, and the delegating party has a duty of supervision over the processor.
For meeting data held on business plans (Team and Enterprise), UNCHAIN, Inc. acts as a party to whom the customer has delegated its handling. Use for training would go beyond the scope of the work delegated, so within that framework we do not use customers' meeting data to train UNCHAIN, Inc.'s own models. That is why there is no use for training on business plans.
Sending data to external model providers
Where data is sent to an external model provider, UNCHAIN, Inc. selects only providers meeting the following conditions.
- it is established by contract that data sent is not used to train their models
- the retention period is stated, and data is not retained beyond the period necessary
- arrangements allow supervision of the processor under Article 25 of the Act on the Protection of Personal Information
Where a model provider is located outside Japan, customers' personal data is transferred abroad. The rules on cross-border transfer apply even where the provision accompanies delegation, so UNCHAIN, Inc. complies with Article 28 of the Act. The destination countries, their legal regimes, the measures taken, and how information is provided on request are set out under “Provision to third parties in foreign countries” in our Privacy policy. For the key points only, see Security and privacy FAQs.
The position under copyright law
Material used in a meeting, and speech such as a lecture or presentation, can contain third parties' copyright works.
This provides that where the purpose is not to enjoy, oneself or through others, the thoughts or sentiments expressed in a work, it may be used to the extent considered necessary. Item (ii) of that Article covers use for “information analysis”, which includes use for machine learning. The Article has a proviso, however: it does not apply where use “would unreasonably prejudice the interests of the copyright holder in light of the nature and purpose of the work and the manner of its use”.
That said, being permitted under copyright law and being permitted under contract and under the Act on the Protection of Personal Information are separate questions. Meeting records contain participants' personal data, business secrets, and information covered by non-disclosure agreements with clients. What governs the handling of those is not copyright law but the Act on the Protection of Personal Information and the contract between the customer and UNCHAIN, Inc..
UNCHAIN, Inc. therefore does not rest the basis for use in training on a general permissive provision of law, whatever the copyright position. On the free and Personal plans the basis is the user's own setting; on business plans we do not use data for training, because it would go beyond the scope of the delegation.
If something is not working
- The item is not in my settings: on business plans (Team and Enterprise) data is not used for training, so the on/off option is not shown at all. You can check your plan under Plans and billing.
- An internal review asks for evidence that data is not used for training: an administrator can check what is displayed under Data security in the workspace settings and use that in their answer. If you need it in writing, speak to your account representative.
- I want a meeting already recorded excluded from training: delete the note and transcript concerned. How to delete is covered in Deleting notes.