At Absorb¹, we are dedicated to developing and deploying generative artificial intelligence (GenAI) solutions that are trustworthy, secure, and responsible, while aligning with our clients' needs and values. Our GenAI Program is designed to align to the NIST AI Risk Management Framework (NIST AI RMF), promoting safety, transparency, and resilience in our GenAI offerings. 

This overview is accurate as of the document's effective date. Because the GenAI landscape is evolving rapidly, including ongoing improvements to models, capabilities, and industry standards, Absorb may update this document and associated program details from time to time to reflect product changes, risk controls, and emerging best practices.

To stay informed of revisions and updates, clients should follow this Knowledge Base (KB) article and subscribe to notifications.

Our GenAI Commitment

Absorb is committed to:

• Advancing trustworthy GenAI that aligns with ethical, legal, and societal expectations.
• Embedding risk management practices throughout the GenAI lifecycle.
• Promoting transparency and explainability in GenAI-driven functionalities.
• Safeguarding users' data privacy and security.
• Championing human-centric GenAI development, with appropriate human oversight.

Absorb's AI Program — How Does Absorb Leverage and Deploy GenAI?

Absorb delivers advanced GenAI-driven services by combining various enterprise-grade service providers of large language models. Leveraging diverse platforms under licenses helps to ensure scalability, security, and compliance for our clients. This approach enables us to mitigate risks effectively, maintain operational resilience, and adapt to evolving regulatory landscapes, all while prioritizing transparency in our processes and outcomes.

I. Architecture and Data Handling

Absorb's GenAI architecture is built on a layered, modular framework integrated within our Learning Management System (LMS), ensuring scalability, robust security, and efficient data handling as further displayed in Chart 1, illustrating the architecture, and Chart 2, listing the GenAI services data handling, each below.

a. Architecture

The architecture is composed of four layers, each with distinct responsibilities and safeguards.

• Top Layer — User Layer: where different user roles interact with GenAI features. Interactions flow through the LMS core platform, which serves as the central hub with components including an API gateway for secure entry, input validation to ensure prompt integrity, a response handler for output processing, and a content manager for overseeing AI-generated materials.
• Layer 2 — Security and Compliance Layer: helps to enforce critical safeguards such as encryption for data in transit, access control to manage permissions, audit logging for traceability, and data privacy measures to protect sensitive information. To support this layer, Absorb stores GenAI inputs and outputs for up to 90 days for auditing and monitoring purposes.
• Layer 3 — AI Services Layer: integrates both third-party AI providers and proprietary AI models to power GenAI functionalities.
• Last Layer — Data Layer: stores and manages information across the LMS database (for user data and client content), AI interaction logs and conversation history (for audit and monitoring), and analytics database (for performance metrics). Key architecture features include encrypted data transmission, API rate limiting and timeouts, comprehensive audit logging, fallback mechanisms, and isolation from the core LMS to enhance resilience.

The data flow is user-initiated as a push mechanism: Users (Learners, Creators, or Administrators) input or upload data through secure interfaces in the user layer, generating and submitting text prompts that incorporate the input information. Absorb does not pull or ingest user data directly into the LLMs; instead, these client-pushed prompts are validated and processed through the layers, with risk assessment aligned to the NIST AI RMF to mitigate biases, hallucinations, or privacy issues. Our workflow enables the client to review outputs prior to deploying or applying them. Outputs are delivered to the client, and the client is responsible for conducting and applying any review, validation, or approvals the client deems necessary before use. This overall workflow further ensures traceability, accountability, and ethical standards while optimizing performance across the decentralized GenAI services.

Chart 1. Architecture

b. Data Handling

For our GenAI features, we partner with select enterprise-grade providers of large language models. These providers may process client inputs submitted in connection with the applicable GenAI feature in order to generate AI-assisted outputs. Client data and content submitted through these features are not used to train the underlying large language models. Below is a summary of the key providers, their purposes, applicable data handling and retention practices, and related rights considerations. This information is based on our contractual arrangements and is subject to periodic review.

Chart 2. GenAI

The following table summarizes the Absorb AI features, their service providers, purposes, data processed, retention periods, and rights to output.

² Absorb leverages Arize for AI observability as part of the "Security and Compliance" layer illustrated in Chart 1. Data processed through Arize for observability purposes is retained for up to 90 days. Any retention by the applicable large language model provider is subject to that provider's retention period as identified in the applicable table or documentation. Separately, prompts, outputs, and conversation history stored by Absorb within the LMS are retained independently of such provider-side and observability retention, including, where applicable, for so long as the client remains a customer of Absorb.

³ To the extent permitted by applicable laws, the client retains its rights, title, and interest in and to client content, and Absorb claims no ownership of client content. AI-generated output does not transfer to the client any ownership rights in any Absorb or third-party-provided materials that may be incorporated in, reflected in, referenced by, or used to generate such output, including templates, prebuilt learning paths, skill frameworks, taxonomies, sequencing logic, content, data, tools, or models. All such Absorb and third-party materials remain the property of Absorb or the applicable third party. The client's rights to use such materials, including as embedded in or necessary to use the AI-generated output, are limited to the rights expressly granted under the Agreement and any applicable third-party terms.

II. Service Descriptions of AI Features

This section describes each Absorb AI feature, including its purpose, scope, and key GenAI-driven capabilities.

a. Create AI

The Absorb Create tool is a GenAI-enabled online Course builder designed to streamline the development, editing, and publication of high-quality educational content. Outputs generated through these features remain under the client's control, and Absorb does not claim ownership of client inputs or outputs through use of the feature. Prompts and related inputs submitted through these features may be processed as necessary to provide the feature.

Key GenAI-driven features include:

Generative Course Builder

This feature structures and designs complete Courses based on user-provided inputs such as topics, objectives, and source materials. It generates outlines, content sections, interactive elements (e.g., animations, buttons, videos), and quizzes, with the goal of supporting alignment with educational goals. Users can refine and customize the output to incorporate branding, themes, or specific pedagogical approaches.

GenAI Quiz Generator

Integrated with the Course Builder, this tool creates quizzes tailored to the Course content. It supports various question types (e.g., multiple-choice, true/false, interactive scenarios) and ensures assessments intended to be effective for measuring Learner outcomes. Quizzes can be generated with minimal input, such as a prompt or content summary, and are fully editable by Users.

GenAI Voiceover Narration

To enhance accessibility and consistency, GenAI-generated voices provide narration for course modules. This feature supports multiple languages and tones, allowing for more consistent audio delivery across courses authored by different team members.

GenAI Generated Images (beta, as available)

Clients may select pre-existing images from an integrated image library. As with other GenAI-generated content, generated images may be duplicative or contain inaccuracies. Clients are responsible for reviewing and verifying outputs for suitability and alignment with intended use, including with respect to applicable intellectual property and other legal requirements.

Podcast

This tool is a GenAI-enabled feature designed to enable clients to generate podcasts. Integrated within Create AI, it transforms user inputs into audio episodes. Outputs generated through Create Podcast remain under the client's control, and Absorb does not claim ownership of client inputs or outputs through use of the feature. Prompts and related inputs submitted through this feature may be processed as necessary to provide the feature. Key GenAI-driven features include:

• Prompt-Based Podcast Generation: Clients provide textual prompts specifying details such as podcast participants (e.g., hosts, guests), voices (e.g., synthetic voice styles or tones), content topics to discuss, overall tone (e.g., formal, conversational, energetic), episode structure (e.g., intro, segments, outro), duration, and other customizations. The GenAI synthesizes these inputs to generate a cohesive audio podcast.
• PDF Upload Integration: For streamlined content sourcing, clients can upload PDF documents (e.g., scripts, articles, or reports), which the GenAI analyzes to extract key themes, summarize content, and generate a podcast episode. This feature combines with prompt-based customizations for refined outputs, such as adapting the narrative to specific audiences or incorporating client branding.

b. Skills

The Absorb Skills tool is a GenAI-enabled upskilling feature within the Absorb LMS. It recommends learning paths by analyzing learner goals, self-assessments, and skill gaps to suggest next steps. GenAI-assisted auto-tagging links relevant skills to uploaded course content, and related reporting helps visualize organizational progress, strengths, and skill gaps. Use of Skills is optional and customer-controlled.

Key GenAI-driven features include:

GenAI-Driven Learning Paths

This feature analyzes individual User profiles, goals, progress, and identified skill gaps to generate personalized learning paths that help tailor learning journeys across available content sources and support continued skills development as roles evolve. These recommendations are intended to help identify potential knowledge gaps and suggest practical next steps. Learners and Administrators remain in control and can review, adjust, or ignore them to fit their workflows. The recommendations do not automatically trigger enrollments, assignments, removals, approvals, or other actions and Users decide whether and how to apply them before taking action.

Skills Auto-Tagging

GenAI analyzes uploaded courses or content and applies relevant skills, enabling rapid integration of proprietary, custom, or third-party materials. This feature maps skills to content, streamlining the creation of comprehensive upskilling frameworks and reducing manual tagging efforts.

c. Aura

The Aura family of features brings GenAI assistance to Administrators, Learners, and Course Authors within the Absorb LMS.

Aura Admin Assist 

Aura Assist is an evolving GenAI companion designed to assist LMS Administrators within our platform. In its current phase, it is grounded in Absorb documentation and related platform resources, retrieve relevant details, and provide step-by-step guidance that enhances administrative efficiency. Aura Assist is optional and enabled at the client's discretion and is currently available exclusively to Administrators. 

In its current phase, where Aura Assist is used to complete administrator-requested tasks, prompts are de-identified before being sent to the large language model. In future roadmap releases of Aura Assist, to expand the available task completions, administrator-requested tasks that require access to personal data within the LMS, will send that personal data to the LLM model, provided that the LLM does not (i) train, fine-tune or improve on the personal data, (ii) retain the personal data beyond the retention period stated in Chart 2 above for the stated purposes, (iii) only has access to the personal data necessary to complete the administrator-requested tasks, (iv) only processes the personal data if so prompted by a client administrator, and (v) the LLM is listed as a data sub-processor subject to vendor management due diligence. Except as described above with respect to de-identified task-completion prompts, administrator prompts and related inputs submitted through Aura Assist, may be processed as necessary to provide the feature. Any content retrieved in response to a given prompt may be included in the input submitted to the LLM model to support generation of a response but it is not used to train the LLM model.

Key GenAI-driven features include:

• Query Answering and Information Retrieval: This feature processes administrator-submitted questions using Absorb documentation, live client data within the LMS, and related platform resources.
• Step-by-Step Guidance: For procedural or troubleshooting matters, this feature generates sequential instructions grounded in Absorb documentation and related support content, and, where enabled by the client, client-selected indexed content. It supports Administrators in navigating LMS workflows, including configuration and troubleshooting scenarios, and may refine its guidance based on follow-up questions or additional context.
• Task Completion Assistance: Aura Assist may also assist Administrators in completing certain tasks within the LMS. For these task-completion requests, prompts are de-identified before being transmitted to the large language model.

Aura Learner Agent

Aura Learner Agent is available within the Learner Experience in the LMS and, where enabled by the client, through supported integrations. Users can ask questions in natural language and get immediate, tailored support, whether they are trying to find a course, decide what to take next, understand a topic covered in training, review transcriptions, or identify outstanding courses. 

Aura Learner Agent is optional to use and can be made available to client users by Administrators. Client Admins may choose which Courses, files, documents, and other supported content are made available to Aura Learner Agent, whether shared directly or through supported integrations. Only content selected by the Client is indexed for this purpose, and responses may be grounded in that indexed content. Client-selected content indexed for use by Aura Learner Agent is not used to train the large language model for this feature. Instead, relevant information may be retrieved from indexed content in response to a given prompt and may be included in the input submitted to the large language model to support generation of a response. Aura Learner Agent provides information and recommendations to support user decision-making, but the user remains the final decision maker and the agent does not make decisions on the user's behalf.

Key GenAI-driven features include:

• Search available courses to help users quickly locate relevant offerings based on role, topic, skill level, or keywords.
• Provide course recommendations by suggesting options that align to the user's stated goals or interests. For course recommendation use cases, recommendations are made based on the learner's available courses.
• Look through course content, and, where enabled by the client, client-selected content, to respond to questions grounded in the material, helping learners get answers and context without needing to manually hunt through modules.
• Provide additional learning support, such as surfacing outstanding courses or providing transcriptions, as available.

By meeting users where they already work and learn, the agent reduces time spent searching, improves discoverability of content, and helps users move from questions to action faster, while leaving decisions regarding course selection and other actions to the user.

Aura Create Agent (beta)

Aura Course Authoring Agent includes an agent embedded directly in Absorb's Admin Experience UI via a chat interface. It supports Admins in building courses more efficiently by translating natural-language prompts into structured learning content, allowing Admins to move from an idea to publish-ready courses with fewer manual steps. Aura Course Authoring Agent is optional to use and made available to Administrators, as applicable. Client-selected content indexed for use by Aura Course Authoring Agent is not used to train the large language model for this feature. Instead, where such content is enabled, relevant information retrieved from indexed content in response to a given prompt may be included in the input submitted to the large language model to support generation of a response. Administrators select which Courses, files, documents, pages, and other supported content are made available to Aura Course Authoring Agent, whether shared directly or through supported integrations. Only content selected by the Administrator is indexed for this purpose.

Key GenAI-driven features include:

• Create Courses for Learners by helping draft Course outlines, module structures, learning objectives, and descriptions based on what the admin is trying to achieve.
• Generate course content (e.g., lesson text, knowledge checks, summaries, and supporting materials) aligned to the Admin's inputs and desired level of detail.
• Use client-selected knowledge sources by retrieving relevant information from files, documents, pages, and other supported content shared with Absorb, whether directly or through supported integrations. Administrators control which content is selected and indexed for this purpose. The agent may use relevant information retrieved from that indexed content to produce course content that reflects the organization's terminology, policies, and internal context, reducing rework and helping maintain consistency.

By working within the Admin workflow and leveraging the content you already maintain elsewhere, the agent helps speed up course creation, improve content reuse, and make it easier to keep learning materials accurate and up to date.

d. Rich Text Editor

The Rich Text Editor tool is a GenAI-enhanced component integrated within the Absorb LMS platform, designed to facilitate the creation, editing, and refinement of textual content. Use of this GenAI feature is optional and within the client's control. Outputs generated through this GenAI feature remain under the client's control and Absorb does not claim ownership of client inputs or outputs through use of the feature.

Key GenAI-driven features include:

GenAI-Assisted Text Editing

Accessed via a dedicated AI button, this feature allows Users to submit text for GenAI-driven modifications. The GenAI can rephrase content for clarity or conciseness, correct grammar and spelling, suggest improvements in tone or structure, summarize lengthy passages, expand on ideas, or adapt text to specific audiences (e.g., simplifying complex language for learners). Users provide prompts or select editing modes to guide the process, ensuring outputs align with intended goals.

e. Together Software

Together Software includes several GenAI-enabled features that support mentorship, profile creation, and meeting preparation.

Profile Bio Generation

Uses GenAI to create personalized professional biographies tailored to Users' backgrounds. Users may select their preferred input method, including uploading a PDF exported from their LinkedIn profile or uploading a resume PDF for analysis. Users retain control over the generated biography and may edit, refine, or regenerate the output as needed to align with their personal voice and preferences.

Personalized Agendas

Uses GenAI to generate tailored session agendas for mentorship pairings or collaborative meetings. Each agenda may be generated based on participant registration profiles, prior session agendas, and any user-supplied talking points intended to address specific goals or topics. This feature produces context-aware outlines designed to improve relevance and productivity. Users may edit the GenAI-generated agendas to add details or make adjustments, maintaining flexibility and human oversight throughout the process.

GenAI Generated Mentorship Program

Uses GenAI to assist Administrators in building structured mentorship programs. Through an interactive question-and-answer interface, the feature gathers information regarding program objectives and other relevant factors and generates a program outline based on those inputs. This feature streamlines program design while allowing Administrators to review, revise, and modify the generated content to align with organizational needs and preferences.

III. Client Controls and Customization

To empower clients and align with the NIST AI RMF's emphasis on governance and manageability, certain AI Features enable clients to:

a. Enablement and Disablement

The GenAI features may be disabled or enabled by the client at their choosing, subject to the applicable feature configuration and availability.

b. Model Selection

Certain of the GenAI features support different LLM providers. Clients may choose to leverage a different LLM by requesting that such provider or model be enabled, where supported by the applicable feature.

c. Customization and Editing

GenAI-generated content is editable, enabling clients to review, modify, or override outputs to ensure accuracy, relevance, and alignment with internal standards. Collaboration tools in certain GenAI features support team-based reviews, task assignments, and feedback loops. Where applicable, clients may also control which courses, files, documents, pages, and other supported content are made available for retrieval-based features, including by selecting content for indexing directly or through supported integrations.

d. Output Ownership

The following provisions describe ownership of GenAI-generated output, the Absorb platform, and any third-party materials referenced in or used to generate that output.

• As between Absorb and the client, the client owns generated output produced for the client to the extent such output is protectable and ownership is recognized under applicable law. This includes learner-facing answers and summaries delivered to the learner, admin-generated draft course materials created by the client, and any course content that an admin chooses to save, publish, or otherwise incorporate into the client's LMS.
• Absorb retains ownership of the Absorb platform and UI, including the admin experience, learner experience, and supported integrations, the agent framework, system prompts, orchestration, retrieval and ranking logic, models and configuration, and any platform features used to produce the output, and pre-existing Absorb content, templates, taxonomies, metadata structures, and product documentation (except where expressly client-owned).
• If output includes or is based on third-party content the client has connected or third-party course content made available through libraries or marketplaces, then the underlying rights in that third-party content remain with the applicable third party. The client's use of that material (and anything derived from it) remains subject to the relevant licenses and terms.
• For clarity, AI-generated output does not transfer to the client any ownership rights in Absorb or third-party-provided materials incorporated in, reflected in, referenced by, or used to generate such output, and all such materials remain subject to the applicable ownership, license, and use restrictions.

IV. Model Training and AI Data Use FAQ

The following frequently asked questions address how Absorb processes, retains, and protects data submitted to GenAI features, and how that data relates to model training.

a. What data does the AI process?

The AI processes only what is needed to provide the requested functionality. The client controls its input content. The LMS does not pull client content into the AI Features without the applicable client prompt or other client-directed use of the feature. Depending on the AI Feature, this can include:

• User Inputs: prompts, questions, instructions, and feedback provided through the applicable feature.
• Absorb Platform Context: information required to fulfill the request (for example, course catalog metadata, learner enrollments and completions, permissions and role context, and other in-product signals as applicable).
• Content the AI is Asked to Reference: course content available in Absorb (for the Learner Agent), and, for course authoring and other retrieval-based features, client-selected files, documents, pages, and other supported content shared with Absorb directly or through supported integrations. Where applicable, such content may be extracted and indexed to support retrieval of relevant information in response to a prompt.

b. Is our data shared with other clients?

No. Client content is not shared across clients or used to generate responses for other clients. AI-generated output based on a client's content is not shared with other clients.

c. Do you use our data to train third-party AI models?

No. We do not use client inputs, client files or content, or AI-generated output to train, fine-tune, or otherwise improve third-party foundation models. We do not provide client content to third parties for model training.

d. Do you train Absorb models on client data or client content?

The following describes how Absorb handles user prompts and course authoring or client-provided source content with respect to model training.

• User prompts: Absorb may use limited interaction data, such as prompts, responses, and quality or feedback signals, to improve Absorb's own AI capabilities, subject to privacy- and security-aligned controls such as data minimization, access controls, and, where appropriate, aggregation and/or de-identification.
• Course authoring and client-provided source content: Absorb does not train its models on client-provided course authoring materials, including uploaded files, connected-source content, or client-specific AI-generated course content, without prior client consent, unless such training is performed solely for that client's benefit and is not reused for other clients.

e. What's the difference between "processing" and "training"?

The terms describe two distinct uses of data, as outlined below.

• Processing means the AI uses data at request time to generate an answer or draft. For example, processing may include analyzing a prompt, using platform context, or retrieving relevant information from indexed content to support a response.
• Training means using data to update or improve a model for future use. Absorb does not use client-provided course authoring materials, connected-source content, or client-specific AI-generated course content to train third-party models, and does not use such materials to train Absorb models except as described above. As described in Section d, Absorb may use limited prompt and interaction data to improve Absorb's own AI capabilities, subject to applicable privacy and security controls.

f. Could our AI output be similar to what other clients see?

Yes. Depending on the prompt, the AI may generate the same or similar responses for different clients, particularly for general product guidance or "how-to" questions that are not client specific. For example, if an admin asks Assist how to enroll a user, the steps and response will often be the same (or substantially similar) across clients because the underlying product functionality is consistent. This similarity does not mean client data is being shared, but it reflects that common questions often have common answers.

g. Does the AI retain or learn from our connected knowledge sources?

Connected sources are accessed only when the client enables them and shares content with Absorb for the applicable feature. Where applicable, client-selected content from connected sources may be extracted and indexed to support retrieval of relevant information in response to a prompt or to support client-directed content generation. Such content is not used to train third-party models and is not used to train Absorb models except as expressly described above.

h. Who can access our prompts, files, and AI outputs?

Access is restricted to authorized users and systems required to operate and support the service, consistent with role-based access controls and internal security policies. Client data is not made available to other clients.

i. How is AI data retained (prompts, outputs, and connected-source content)?

Retention is governed by the practices outlined below, which describe LLM provider retention, Absorb's retention within the LMS, and retention of client-selected indexed content.

• The applicable LLM provider processes prompts and related inputs to generate a response and may retain such data for a limited period in accordance with its applicable security, monitoring, and auditing practices, as described in Chart 2 above. Client data and content submitted through the AI Features are not used to train the underlying third-party models.
• Separately from any retention by the applicable LLM provider or AI observability tool, Absorb retains conversation history, prompts, and AI-generated outputs within the LMS or associated application records for the term of the agreement in order to support continuity of service, including troubleshooting, support, and customer experience continuity. Where such data is maintained in de-identified and aggregated form, it is no longer associated with an identifiable client or individual.
• Where client-selected content is made available for retrieval-based features, such content may be extracted and indexed to support retrieval of relevant information in response to prompts. Indexed content is retained in accordance with the client's configuration and the applicable service lifecycle.