The Glean AI Addendum supplements the Terms of Service for any service that uses generative AI. It sets out how Glean manages the way Customer Data is collected, permission-filtered and sent to models, and how third-party model providers handle that data under their own security, safety and acceptable-use terms.[1] Glean positions the service as an open platform that supports models from providers such as OpenAI, Anthropic and Google through a common interface, so that customers can choose or change models by use case, cost or regulatory need.[1] For licensing purposes the question is which model key the customer uses, because the key determines which commitments apply. The overview is Glean licensing, and pricing for model use is in Glean Enterprise Flex, Core Suite and FlexCredits.
Editions
The AI terms divide customers by who holds the model key.
| Option | Who supplies the model access | Commitments that apply | Source |
|---|---|---|---|
| Glean’s model key (the Glean Universal Model Key) | Glean | No training on Customer Data by Glean or the providers, zero-day retention, Output IP indemnity | AI Addendum[1] |
| Customer’s own model key (Customer Key) | The customer | Glean does not train on Customer Data; the customer bears sole responsibility and liability for its model access | AI Addendum[1] |
| Limited Retention Model | Glean or customer key, with a model designated as not zero-retention compliant | Retention only as required by law or for trust and safety, with no training | Limited Retention Addendum[2] |
The same distinction affects price. Customers using a Customer Key keep a different Deep Research and agent-run treatment, Model Hub Usage is limited to the Glean Model Family for customers on a Customer Key, and seats and FlexCredits are discounted for customers that supply their own LLM keys.[3][4]
Metrics
The AI terms do not define a licensed unit. Consumption is measured by the pricing documents. Model use is drawn as Model Hub Usage under Core Suite or FlexCredits under Enterprise Flex, and the AI terms add two definitions that matter for ownership: Input is any natural language statement, SQL request, prompt or other query a User provides to the service, and Output is any generative AI output provided by a model through the service after processing an Input.[1] Under the Terms of Service the customer retains all rights in Customer Data, including Input and Output, and Glean receives a right to process it to provide the service or address technical problems during the Subscription Term.[5]
Counting / floors
Glean’s commitments
The AI Addendum lists these commitments.
- Certified management system. Glean maintains an artificial intelligence management system certified to ISO/IEC 42001:2023 by an independent accredited body for its enterprise AI solutions, and must maintain it or a substantially equivalent successor during the Subscription Term, operate in accordance with it, and undergo periodic recertification audits.[1]
- No training. Glean will not use Customer Data to train, re-train, improve or fine-tune any AI models. For the Glean model key it keeps contractual commitments that bar model providers from doing so.[1]
- Zero-day retention. For a service using Glean’s model key, Glean will implement zero-day retention, including through its commitments with third-party model providers, so that they do not store or log Customer Input or Output, except as required by law or as necessary for trust and safety such as enforcement of acceptable use policies.[1]
- Compliance with law. Glean provides the service in compliance with laws applicable to its provision of the service generally, including the EU AI Act, without regard to the customer’s particular use.[1]
Output indemnification
In addition to the general indemnity for infringement claims, Glean’s duty to defend and indemnify against third-party intellectual property claims extends to each Output, but only for a service using Glean’s model key and only on conditions. The customer must have used the service in compliance with the Agreement. The Output must not come from a model that the customer trained, fine-tuned or provided. The customer must not have disabled or evaded content filters or safety systems, modified the Output, solicited infringing Output through its Input, or lacked sufficient rights in the data and Input. Nor may it have used or distributed the Output in a way it knew or should have known would violate a third party’s intellectual property rights.[1] A customer using its own key therefore does not have this extra indemnity from Glean, and it is solely responsible and liable to the extent it accesses models with its own key.[1]
Limited Retention Addendum
An administrator who enables, through the admin console, a supported model designated as non-compliant with zero data or zero-day retention is deemed to accept the Limited Retention Addendum for that model. Under it, Glean or the model provider may retain and process Customer Data, including Input and Output, only to the extent required by law or necessary for trust and safety. Neither Glean nor the providers may use the data to train or improve any model.[2] The consequence is that model enablement is a contractual act. An admin who turns on a limited-retention model changes the data terms for the whole organisation without a separate signature.
Customer commitments
The customer is responsible for the lawfulness of all Input and for decisions, actions and inactions arising from its use of AI features, including compliance with data protection, intellectual property and AI laws. It understands that Output may be inaccurate, inappropriate or biased and must implement reasonable practices such as human oversight, and it must not use AI features in violation of the Acceptable Use Policy.[1]
Acceptable use of AI features
The Acceptable Use Policy adds AI-specific limits in section 2. The customer may not use AI features to make Fully Automated Decisions that produce legal or similarly significant effects on individuals, such as in finance, employment, healthcare, housing, insurance or social welfare, unless the decisions comply with law with human review and final decision making and the customer is transparent about the role of AI and the main reasons for the decision.[6] It may not disable, evade or disrupt safeguards, misrepresent Output as human-generated, or remove or alter indicators that Output is AI-generated.[6] The AI features are not intended for prohibited practices or high-risk uses under the EU AI Act or for sensitive, critical, unsafe or hazardous uses, and the customer is responsible for appropriate AI disclosures to consumers.[6] The customer may not use the service, AI features or Output to train, re-train, fine-tune, test, improve or enhance products or services that compete with the service, including other AI models.[6] If the Acceptable Use Policy conflicts with the Agreement, the Policy governs.[6]
Third-party models
The Optional Offerings document lists third-party developed or hosted models and the licence agreements and use policies that apply when a customer elects them. The list covers models offered through Amazon Bedrock, Azure OpenAI, Google Vertex, Meta Llama, Anthropic, OpenAI and DeepSeek, with a licence agreement and a use policy for each platform. The customer must comply with the provider’s terms, which supplement the Agreement or form an independent agreement with the customer, and which may be updated by the provider.[7] The Terms of Service add that access to and use of models is governed by their respective terms and the Agreement.[5] Because the model catalogue changes often, the Optional Offerings document is an inventory that a software asset manager should recheck at renewal, and the Core Suite and Enterprise Flex pages carry the current model lists and tiers.[4][3]
Virtualization & partitioning
Not applicable. The partition is the model key: Glean’s key and the customer’s key carry different commitments, and enabling a limited-retention model changes the data terms for the organisation.
Cloud / BYOL
The customer may host in its own cloud account and bring its own model key, which Glean describes as a choice that reduces Enterprise Flex pricing. In that case the customer implements Localized Updates, and Glean is not liable to the extent a failure to do so causes an issue.[5][3]
Programs
- Customer Key and Glean Universal Model Key. Model key options.[1]
- Limited Retention Addendum. Accepted by enabling a Limited Retention Model.[2]
- Trials and Preview Features. Previews are excluded from security, compliance and privacy commitments and should not process personal data.[8]
Out of scope
- The Security and Privacy Standard and audit reports, which describe controls rather than licence terms.
- Model providers’ own enterprise agreements where the customer brings a key.
- Contractual position under national AI laws other than those named in the Glean documents.