AI Addendum

Capitalized terms not defined here have the meaning given in Section Definitions or, where applicable, the executed Master Services Agreement.

1. Scope

This AI Addendum applies to Customer's use of AI Features within the Services. It anchors Section's published positions on architecture, training, ownership, accuracy, bias, and Foundation Model selection. The MSA governs commercial terms, and the Data Processing Addendum governs the processing of Personal Data. This addendum does not duplicate or alter those terms.

2. Local definitions

In addition to terms defined in the Definitions document, the following local terms apply:

Inputs: content (prompts, files, instructions, or other materials) submitted by Customer or an End User to the AI Features.

System Prompt: instructions configured by Section that direct a Foundation Model's behavior within a defined task domain.

3. Architecture

Section, Inc. ("Section", "we", "us", "our") operates AI Features as a wrapper and retrieval-augmented layer on top of third-party Foundation Models. We do not train, fine-tune, or host any Foundation Model. We invoke Foundation Models server-to-server through the LLM gateway under contracts that prohibit retention of Inputs or Outputs and prohibit use of Inputs or Outputs for model training.

The current Foundation Model providers and LLM gateway are listed on the Subprocessor List.

4. No model training on Customer Data

Section will not use Customer Data to train, fine-tune, or evaluate any Foundation Model. Our contracts with the LLM gateway and Foundation Model providers carry the same prohibition.

5. Output ownership

As between Section and Customer, Customer owns the Outputs generated for it through use of the AI Features, subject to: (i) Section's underlying intellectual property in the AI Features, the System Prompts, and the Services; (ii) common-law and statutory limits on the protectability of AI-generated content under applicable intellectual property law; and (iii) the Acceptable Use Policy.

Outputs are provided "as is." Customer is responsible for reviewing Outputs and verifying that any Output does not infringe a third party's intellectual property, publicity, privacy, or other rights before reproducing, distributing, publishing, or otherwise using the Output for a commercial purpose.

6. Accuracy

The AI Features produce probabilistic Outputs. Section does not warrant that any Output is accurate, complete, current, or fit for any particular purpose. Outputs may contain errors, omissions, or fabrications ("hallucinations") notwithstanding the controls described in section 7.

7. Accuracy and content controls (process, not warranty)

Section operates the following controls to reduce, but not eliminate, accuracy and content-safety risks in Outputs:

1. Constrained application design: AI Features are scoped to defined task domains (coaching, certification, use case discovery, analytics) through purpose-built System Prompts. The AI Features are not an open-ended chatbot.
2. Grounding: prompts incorporate Customer-specific or role-specific context to ground Outputs in verifiable information where applicable.
3. Parameter tuning: temperature and related parameters are tuned to favor grounded, lower-variance responses over speculative ones.
4. Input validation: Inputs are validated to mitigate prompt injection and similar manipulation.
5. Output filtering: Outputs are monitored for inappropriate or harmful content.
6. Human-in-the-loop: Outputs are presented as recommendations for human review, not autonomous decisions. End Users determine whether to adopt, modify, or discard a recommendation.

These controls are documented further in Section's GenAI Technical Addendum, available to Customer under NDA.

8. Bias mitigation (process, not warranty)

Section operates the following controls to reduce, but not eliminate, bias risks in Outputs:

1. Foundation Model selection: LLM Providers are selected based in part on published responsible-AI and bias-mitigation frameworks.
2. System Prompt design: prompts emphasize objectivity, inclusivity, and relevance to each user's role and context, and scope Outputs to defined task domains to reduce the surface area for biased generation.
3. Monitoring: Outputs are reviewed periodically across user populations and roles, and recommendation patterns are analyzed.
4. User reporting: End Users may report problematic Outputs through the in-product reporting mechanism.
5. Governance: AI governance is aligned with ISO/IEC 42001 (see section 12), and all personnel complete GenAI awareness training.

Customer acknowledges that bias cannot be fully eliminated from probabilistic systems and that Section provides these controls on a reasonable-efforts basis, not as a warranty of unbiased Outputs.

9. Customer responsibilities

Customer is responsible for:

1. Human review for high-stakes decisions. Customer must not use an Output as the sole basis for any decision concerning employment, compensation, discipline, termination, eligibility for benefits, legal status, immigration status, medical care, financial credit, insurance, housing, or other consequential matters. A qualified human must review the Output and exercise independent judgment.
2. Data scope. Customer must not submit, and must instruct its Authorized Users not to submit, protected health information, government-issued identifiers, financial account data, biometric data, or sensitive personal data through the Services. The Services are scoped to first name, last name, business email, authentication metadata, product event metadata, and (for SectionHQ tenants using the optional HRIS integration) employee directory metadata.
3. Compliance with the AUP. Customer and its Authorized Users use the AI Features in accordance with the Acceptable Use Policy.
4. Verification of Outputs. Before reproducing, distributing, publishing, or otherwise using an Output for a commercial purpose, Customer is responsible for verifying that the Output is accurate and that its use does not infringe a third party's rights.

10. Foundation Model changes

Section selects, replaces, or removes Foundation Models and LLM Providers from time to time as part of operating the Services. Section may make such changes without advance notice to Customer, provided that the change does not materially decrease the functionality of the AI Features and does not weaken the security or zero-retention posture described in section 3 or in the DPA. Material changes to the categories of Sub-processors are governed by section 6 of the DPA.

11. Sub-processors

The LLM gateway and Foundation Model providers are listed on the Subprocessor List. All other terms governing Sub-processors, including notice and remedies, are set out in section 6 of the DPA.

12. ISO/IEC 42001 alignment

Section's AI management practices are designed to align with ISO/IEC 42001 (AI Management System). Section is not certified to ISO/IEC 42001. Alignment is verified through Section's internal control program and is described in the GenAI Technical Addendum and the Section Compliance Playbook, available to Customer under NDA.

13. No expansion of liability

This addendum does not create new indemnities, liability caps, warranties, or remedies. All such terms are governed by the MSA. Nothing in this addendum expands, duplicates, or limits the caps and exclusions set out in the MSA.

14. Order of precedence

In any conflict between this addendum and the MSA, the MSA controls. In any conflict between this addendum and the DPA on data protection matters, the DPA controls. In any conflict between this addendum and the AUP, this addendum controls.

15. Regulatory changes

U.S. and other regulations affecting artificial intelligence are evolving. The U.S. Department of Commerce Bureau of Industry and Security has rescinded the AI Diffusion Framework Rule and may issue a replacement rule. Congress is considering the Remote Access Security Act, which would expand U.S. export controls over remote access to U.S.-origin items. The U.S. Department of Justice operates the Data Security Program under 28 CFR Part 202 governing data transactions involving countries of concern. Section monitors these developments and may modify the AI Features or the Services as needed to comply with applicable law, including by restricting access or features in particular jurisdictions or for particular Authorized Users. A modification made under this section to comply with applicable law is not a breach of this addendum or the MSA.

Section's export-controls representations and warranties, and Customer's corresponding obligations, are set out in the MSA, the Website Terms of Use, the Platform Terms of Use, and the Acceptable Use Policy. This addendum does not duplicate or alter those obligations.

Section, Inc. 228 Park Avenue S, PMB 96268 New York, NY 10003-1502 legal@sectionai.com