Latens

Terms of Service - AI Inference Services

Last updated: 21 May 2026

Contents

• Part 1 - Terms of Service
• 1. Agreement and Scope
• 2. Definitions
• 3. Service Description
• 4. B2B Eligibility, Authority, and Customer Responsibility
• 5. Accounts, API Keys, and Access Controls
• 6. Customer Content and Output Responsibility
• 7. Zero Data Retention and Data Use
• 8. Model Terms and Model Behavior
• 9. Acceptable Use and Regulated Use
• 10. Fees, Billing, Taxes, and Credits
• 11. Service Levels and Support
• 12. Security
• 13. Confidentiality
• 14. Intellectual Property and Feedback
• 15. Disclaimers
• 16. Indemnification
• 17. Limitation of Liability
• 18. Suspension and Termination
• 19. Compliance With Laws and Export Controls
• 20. Changes to These Terms
• 21. Governing Law and Disputes
• 22. Contact
• Appendix A - Data Processing Addendum
• Appendix B - Security Measures
• Appendix C - Acceptable Use Policy
• Appendix D - Service Commitments
• Appendix E - Model Terms and Data Handling Summary

Part 1 - Terms of Service

1. Agreement and Scope

These Terms of Service (the "Terms") are entered into between the Customer and Latens ("Latens", "we", "us", or "our").

These Terms govern access to and use of Latens hosted AI inference services, including API access, dedicated inference deployments, streaming completions, tokenization, account administration, billing, documentation, and support (collectively, the "Services").

The Services are business-to-business services only. The Services are not offered to consumers, children, or individuals acting primarily for personal, family, or household purposes.

If Latens and Customer sign a separate order form, master services agreement, data processing agreement, service level agreement, or enterprise agreement, the signed agreement controls to the extent it conflicts with these Terms.

2. Definitions

"Account" means a Latens account, workspace, organization, API key, billing profile, dashboard access, or other access credential issued by Latens or made available for use with the Services.

"Customer" means the legal entity using the Services, including any organization on whose behalf an individual uses the Services.

"Input" means prompts, messages, parameters, files, system prompts, metadata included in the request body, and other content submitted to the Services for processing.

"Output" means model-generated text, code, data, tool-call payloads, or other content returned by the Services in response to Input.

"Customer Content" means Inputs and Outputs together.

"Usage Metadata" means operational, billing, security, and account data that does not include prompt or completion text, such as timestamp, account identifier, API key identifier, model identifier, endpoint, token counts, latency, status code, error category, request size, and charge amount.

"Model Terms" means license terms, attribution terms, acceptable use terms, usage restrictions, and other legal terms that apply to open-weight, open-source, or other third-party models made available through the Services.

"Zero Data Retention" or "ZDR" means that production inference endpoints process Inputs and Outputs only for the time necessary to provide the inference response and do not intentionally write prompt or completion text to persistent logs, databases, analytics, monitoring systems, backups, or training datasets.

3. Service Description

Latens provides hosted AI inference for open-weight, open-source, and other third-party large language models on dedicated GPU infrastructure in the deployment region agreed with the Customer.

The Services may include dedicated production deployments, model inference APIs, OpenAI-compatible API endpoints, streaming responses, token counting, account management, billing, support, and related tooling.

Latens may add, remove, suspend, rename, replace, quantize, patch, update, or deprecate models, endpoints, parameters, rate limits, context windows, pricing, or features. Latens will use reasonable efforts to avoid material disruption for dedicated production customers.

Latens does not provide legal, medical, financial, tax, accounting, cybersecurity, or other regulated professional advice. Outputs may be inaccurate, incomplete, offensive, unsafe, or unsuitable for Customer use cases. Customer must independently evaluate Outputs before relying on them.

4. B2B Eligibility, Authority, and Customer Responsibility

Customer must be a business, organization, sole trader, professional user, developer acting in a business capacity, or other non-consumer legal or professional user.

Each individual accepting these Terms or using an Account represents that they are at least 18 years old, are not using the Services as a consumer, and are authorized to bind the Customer.

Customer is responsible for its users, applications, API keys, requests, Inputs, Outputs, use cases, end users, downstream distribution, and compliance with applicable laws and these Terms.

Customer must not allow minors to use the Services directly and must not deploy the Services as a child-directed service or consumer-facing general public product unless separately agreed in writing.

5. Accounts, API Keys, and Access Controls

Customer must provide accurate account, billing, tax, and contact information and keep that information current.

Customer is responsible for safeguarding API keys, credentials, dashboard access, and other authentication methods. API keys are secret credentials and must not be embedded in public repositories, client-side code, browser applications, mobile apps, or other environments where they can be extracted by unauthorized parties.

Customer must promptly notify Latens of any suspected unauthorized access, credential compromise, or security incident involving the Services.

Latens may revoke, rotate, rate limit, or suspend API keys where necessary to protect the Services, prevent abuse, comply with law, or enforce these Terms.

6. Customer Content and Output Responsibility

As between Latens and Customer, Customer owns its Inputs. Subject to these Terms and applicable Model Terms, Latens assigns to Customer any rights Latens may have in Outputs generated for Customer through the Services.

Customer is solely responsible for Customer Content, including whether Inputs may lawfully be submitted, whether Outputs may lawfully be used, and whether Customer has all rights, notices, consents, and legal bases required for its use cases.

Customer must not submit content that Customer is not permitted to process through the Services. Customer must not submit special-category personal data, protected health information, payment card data, biometric identifiers, government identification data, customer secrets, classified information, export-controlled data, or other highly sensitive data unless Latens has expressly agreed to that use in a signed agreement.

Latens does not guarantee that Outputs are unique, non-infringing, correct, secure, or fit for a particular purpose. Similar or identical Outputs may be generated for other customers.

7. Zero Data Retention and Data Use

Latens production inference endpoints are operated under ZDR unless a separate signed agreement expressly states otherwise.

Latens does not intentionally store prompt text, completion text, or persistent result caches from production inference requests after request completion. Inputs and Outputs are processed transiently to provide the response and are not used to train, fine-tune, evaluate, benchmark, or improve models or services without Customer opt-in or a signed agreement.

Latens may process Usage Metadata for billing, fraud prevention, rate limiting, security, debugging, capacity planning, service reliability, SLA measurement, and compliance. Usage Metadata must not include prompt or completion text.

Temporary processing required for streaming, queueing, non-persistent in-request KV-cache reuse, retry handling, and network transmission is not considered retention if the data is not intentionally written to persistent storage and is deleted or overwritten as part of normal request handling.

Customer acknowledges that ZDR may limit Latens ability to reproduce requests, debug output quality, investigate model behavior, or restore Customer Content.

8. Model Terms and Model Behavior

Models made available through the Services may be subject to Model Terms imposed by model licensors or publishers. Customer must comply with all applicable Model Terms for the models it uses.

Latens may publish or otherwise make available model information, including model name, model source, license type, context window, deployment region, known restrictions, and support status. Model information may change over time.

Customer must not use a model in a way that violates the applicable Model Terms, including restrictions on unlawful use, regulated use, biometric use, deception, disallowed automated decision-making, or attribution obligations.

Unless expressly agreed in writing, Latens does not fine-tune models on Customer Content and does not create Customer-specific model weights from Customer Content.

9. Acceptable Use and Regulated Use

Customer must comply with the Acceptable Use Policy in Appendix C and all applicable laws, regulations, sanctions, export controls, and industry rules.

High-risk regulated uses are prohibited unless Latens expressly authorizes the use case in a signed enterprise agreement. This includes uses that may require conformity assessment, human oversight, risk management, auditability, or other regulated AI controls under applicable law.

Customer must not use the Services to make or materially assist decisions that produce legal or similarly significant effects for individuals without independently satisfying all legal requirements and obtaining Latens written authorization where required by these Terms.

Latens may suspend or restrict access where it reasonably believes Customer use violates these Terms, creates security or legal risk, threatens service integrity, or may harm third parties.

10. Fees, Billing, Taxes, and Credits

Customer must pay all fees for the Services according to the pricing presented at checkout, in an order form, in an invoice, or otherwise agreed in writing. Fees are usage-based, commitment-based, deployment-based, or a combination of those methods.

Fees are stated in the currency shown at checkout, in the applicable order form, or on the invoice. If no currency is shown, fees are stated in United States dollars. Fees are exclusive of VAT, sales tax, withholding tax, duties, and similar taxes unless expressly stated otherwise.

For self-service prepaid usage, Customer must maintain a positive credit balance unless Latens has approved invoiced billing. Credits are deducted based on actual usage and may decrease in real time. If the balance reaches zero or a risk threshold set by Latens, Latens may suspend or throttle access until additional credits are added or payment is made.

Purchased prepaid credits do not expire under these Terms unless a separate order form, promotion, or applicable law provides otherwise. Promotional, trial, or free credits may expire, may be revoked, and have no cash value.

Prepaid credits are non-refundable except where required by law, expressly stated in a signed order form, or approved by Latens in writing for a verified billing error. Credits are not bank deposits, stored value, or electronic money.

For approved monthly invoicing, Latens may invoice monthly in arrears or according to the applicable order form. Invoices are due thirty (30) days from invoice date unless the invoice or order form states a different payment term.

Customer must raise billing disputes within thirty (30) days after the invoice or charge date and must pay undisputed amounts on time. Past-due undisputed amounts may result in suspension, collection activity, and late payment charges at the lower of 1.5% per month or the maximum rate permitted by law.

Customer is responsible for chargebacks, payment reversals, failed payments, and taxes arising from Customer use. Latens may use third-party payment processors and billing tools to collect payments and manage invoices.

11. Service Levels and Support

The service commitments in Appendix D apply to covered dedicated production inference deployments, unless a signed agreement states otherwise.

Latens targets a 99.95% monthly availability SLO for covered dedicated production inference deployments and a six (6) hour initial response Support SLA for covered production incidents submitted through the designated support contact.

Support response targets are initial response targets, not resolution commitments. The availability SLO is an operational target unless a signed agreement states otherwise. Latens may prioritize incidents based on severity, security risk, production impact, and Customer contract terms.

12. Security

Latens will maintain administrative, technical, and organizational safeguards designed to protect the Services and Customer Content. Current measures are summarized in Appendix B.

Customer is responsible for securing its own applications, systems, networks, devices, API keys, users, end-user notices, and downstream processing. Customer must not attempt to bypass rate limits, authentication, security controls, usage limits, or isolation controls.

Latens may investigate suspected abuse, security incidents, fraud, or policy violations. Investigations will be conducted in a manner consistent with the ZDR commitments in these Terms where technically feasible.

13. Confidentiality

Each party may receive non-public business, technical, security, pricing, product, or operational information from the other party ("Confidential Information"). Customer Content is Customer Confidential Information.

The receiving party must use Confidential Information only to perform or receive the Services, protect it using reasonable care, and disclose it only to personnel, affiliates, contractors, professional advisers, and subprocessors who need to know it and are bound by confidentiality obligations.

Confidentiality obligations do not apply to information that is public without breach, independently developed, lawfully received from a third party without confidentiality restrictions, or required to be disclosed by law after reasonable notice where permitted.

14. Intellectual Property and Feedback

Latens and its licensors retain all rights in the Services, software, documentation, infrastructure, APIs, dashboards, model serving stack, names, logos, trademarks, and related technology.

Customer grants Latens a limited right to process Customer Content solely to provide, secure, bill, troubleshoot, and support the Services and to comply with law and these Terms.

If Customer provides feedback, suggestions, or ideas, Latens may use them without restriction or compensation, provided that Latens does not disclose Customer Confidential Information or Customer Content in violation of these Terms.

15. Disclaimers

Except as expressly stated in these Terms or a signed agreement, the Services are provided "as is" and "as available". Latens disclaims all warranties, whether express, implied, statutory, or otherwise, including warranties of merchantability, fitness for a particular purpose, title, non-infringement, accuracy, availability, and error-free operation.

Latens does not warrant that any model or Output will be accurate, safe, complete, uninterrupted, secure, available in every region, compliant with Customer-specific laws, or suitable for a regulated use case.

Customer acknowledges that AI systems can produce incorrect, biased, harmful, unsafe, or unexpected Outputs and that Customer is responsible for appropriate human review, testing, monitoring, and safeguards.

16. Indemnification

Customer will defend, indemnify, and hold harmless Latens and its directors, officers, employees, contractors, and affiliates from and against claims, damages, liabilities, penalties, fines, costs, and expenses, including reasonable legal fees, arising from Customer Content, Customer applications, Customer use of the Services, Customer breach of these Terms, or Customer violation of law or third-party rights.

Latens will promptly notify Customer of any indemnified claim, allow Customer reasonable control of the defense, and reasonably cooperate at Customer expense. Customer may not settle a claim in a way that admits fault by Latens or imposes non-monetary obligations on Latens without Latens prior written consent.

17. Limitation of Liability

To the maximum extent permitted by law, neither party will be liable for indirect, incidental, special, consequential, exemplary, punitive, or lost-profit damages, or for loss of data, goodwill, business opportunity, revenue, or anticipated savings, even if advised of the possibility of such damages.

To the maximum extent permitted by law, each party total aggregate liability arising out of or relating to the Services or these Terms will not exceed the amounts paid or payable by Customer to Latens for the Services giving rise to the claim during the period before the event giving rise to liability as stated in the applicable order form or signed agreement.

The liability cap does not limit Customer payment obligations, Customer indemnification obligations, either party liability for fraud or intentional misconduct, or any liability that cannot be limited under applicable law.

18. Suspension and Termination

Latens may suspend or limit access immediately if Customer breaches these Terms, fails to pay undisputed amounts, creates security or legal risk, exceeds usage limits, attacks or disrupts the Services, or uses the Services in a prohibited manner.

Either party may terminate for material breach if the breach is not cured within thirty (30) days after written notice, except that Latens may terminate immediately for severe abuse, unlawful use, security threats, or repeated violations.

Upon termination, Customer must stop using the Services and pay all outstanding amounts. Sections that by their nature should survive will survive, including confidentiality, payment, data protection, disclaimers, indemnification, limitations of liability, governing law, and dispute provisions.

19. Compliance With Laws and Export Controls

Customer must comply with all applicable laws and regulations, including privacy, data protection, AI governance, consumer protection, cybersecurity, sanctions, anti-bribery, export control, and industry-specific laws.

Customer must not access or use the Services from embargoed jurisdictions, by sanctioned persons, or for prohibited end uses. Customer must not export, re-export, transfer, or make available the Services or Outputs in violation of applicable export control or sanctions laws.

20. Changes to These Terms

Latens may update these Terms from time to time. The updated version will be effective when posted or otherwise made available, unless a later effective date is stated.

Material changes will not apply retroactively to claims or disputes that arose before the effective date of the updated Terms. Continued use of the Services after the effective date constitutes acceptance of the updated Terms.

21. Governing Law and Disputes

These Terms and any dispute arising out of or relating to the Services are governed by the laws specified in the applicable signed agreement, or otherwise by the laws of the jurisdiction in which Latens is established.

The courts of the jurisdiction in which Latens is established have exclusive jurisdiction over disputes arising out of or relating to these Terms or the Services, unless applicable mandatory law requires another forum or a signed agreement specifies otherwise.

The English language version of these Terms controls. Any translation is provided for convenience only.

22. Contact

Legal, privacy, support, security, abuse notices and business inquiries may be sent to contact@latens-ai.com.

Appendix A - Data Processing Addendum

A1. Scope and Roles

This Data Processing Addendum ("DPA") forms part of the Terms where Latens processes personal data on behalf of Customer in the course of providing the Services.

For Customer Content processed through production inference endpoints, Customer is the controller or processor, as applicable, and Latens is a processor or subprocessor acting on Customer instructions.

For account administration, billing, legal compliance, security, fraud prevention, website, and business contact data, Latens acts as an independent controller as described in the Latens Privacy Policy.

A2. Processing Details

Subject matter: provision, security, support, billing, and administration of hosted AI inference services.

Duration: the term of the Services plus the period required for deletion, backup rotation, legal retention, dispute handling, and audit records.

Nature and purpose: transient inference processing, authentication, rate limiting, billing, security monitoring, incident response, troubleshooting, support, and compliance.

Types of personal data: personal data included by Customer in Inputs; business contact data; account identifiers; usage and billing metadata; support communications; security logs.

Categories of data subjects: Customer personnel, authorized users, Customer end users, and other persons whose personal data Customer submits to the Services.

Sensitive data: prohibited unless expressly authorized in a signed agreement.

A3. Customer Instructions

Latens will process personal data only on documented Customer instructions, including these Terms, an order form, Customer API requests, Customer configuration, and written instructions accepted by Latens.

Latens will promptly inform Customer if, in Latens reasonable opinion, an instruction infringes applicable data protection law, unless prohibited by law from doing so.

A4. Confidentiality and Security

Latens will ensure that persons authorized to process personal data are subject to confidentiality obligations.

Latens will maintain appropriate technical and organizational measures designed to protect personal data against accidental or unlawful destruction, loss, alteration, unauthorized disclosure, or access. The measures are summarized in Appendix B.

A5. Subprocessors

Customer gives Latens general written authorization to engage subprocessors as necessary to provide, secure, support, bill, and administer the Services.

Latens will impose written obligations on subprocessors that are no less protective than the obligations in this DPA, to the extent applicable to the subprocessor services.

Latens remains responsible for subprocessor performance of data protection obligations as required by applicable data protection law.

Where required by applicable law or a signed enterprise agreement, Latens will provide reasonable information about subprocessors upon request and will allow Customer to object to a new subprocessor on reasonable data protection grounds.

A6. International Transfers

Latens supports production inference processing across agreed deployment regions, confirmed during onboarding.

If personal data is transferred outside the European Economic Area, Latens will use a lawful transfer mechanism where required, such as an adequacy decision, Standard Contractual Clauses, or another mechanism permitted by applicable data protection law.

A7. Data Subject Requests and Assistance

Taking into account the nature of processing, Latens will provide reasonable assistance to Customer in responding to data subject rights requests, security obligations, breach notifications, data protection impact assessments, and regulator consultations, to the extent required by applicable data protection law.

Because production inference endpoints operate under ZDR, Latens may be technically unable to access, retrieve, correct, or delete prompt or completion text after a request is complete.

A8. Personal Data Breach

Latens will notify Customer without undue delay after becoming aware of a personal data breach affecting Customer personal data processed by Latens as processor.

The notice will include information reasonably available to Latens to help Customer meet its own legal obligations. Latens may provide information in phases as it becomes available.

A9. Audit Information

Latens will make available information reasonably necessary to demonstrate compliance with this DPA, subject to confidentiality, security, and reasonable scope limitations.

On-site audits require a separate written agreement, reasonable advance notice, normal business hours, appropriate confidentiality obligations, and controls to protect the security and confidentiality of Latens systems and other customers.

A10. Deletion and Return

At the end of the Services, Latens will delete or return Customer personal data processed as processor, unless retention is required by law, dispute preservation, security, backup rotation, or legitimate business records.

Prompt and completion text processed under ZDR is not intentionally retained after completion of the request and therefore generally cannot be returned.

A11. U.S. State Privacy Laws

To the extent U.S. state privacy laws apply, Latens will process personal information as a processor or service contractor for Customer Content and will not sell Customer Content or share Customer Content for cross-context behavioral advertising.

Latens will not retain, use, or disclose Customer Content for purposes outside the business purposes described in these Terms except as permitted by applicable law or Customer written instructions.

Appendix B - Security Measures

B1. Governance and Access Control

Latens maintains role-based access controls, administrative access restrictions, least-privilege access, and administrative multi-factor authentication for systems used to operate the Services.

Latens reviews access to production systems periodically and removes access when it is no longer required.

Latens personnel and contractors with access to confidential information are subject to confidentiality obligations.

B2. Encryption and Key Handling

The Services use encrypted transport for customer-facing API traffic. Latens is designed to use encryption at rest for production databases, backups, and storage systems where supported by the underlying platform.

API keys are stored in hashed or otherwise non-recoverable form where technically feasible. Latens will not intentionally display full secret API keys after creation.

B3. Logging, Monitoring, and ZDR Controls

Operational logs, security logs, billing records, and monitoring systems are configured not to intentionally capture prompt or completion text from production inference requests.

Latens may collect Usage Metadata for billing, rate limiting, reliability, fraud prevention, security, debugging, and SLA measurement.

Debug logging that could capture prompt or completion text is prohibited in production unless expressly authorized for a specific incident or enterprise support case and governed by written Customer instructions.

B4. Availability and Backup

Latens uses operational measures designed to support service resilience, including monitoring, capacity planning, backup processes for business systems, and incident response procedures.

Backups are rotated on a limited schedule and are not designed to restore Customer prompt or completion text processed under ZDR.

B5. Secure Operations

Latens maintains procedures for vulnerability management, patching, incident triage, abuse handling, credential rotation, and vendor review appropriate to the size, nature, and risk of the Services.

Customers may request reasonable security questionnaires for enterprise diligence. Responses may be subject to confidentiality obligations.

Appendix C - Acceptable Use Policy

C1. Prohibited Uses

Customer must not use the Services to violate law, infringe rights, deceive people, harm people or property, compromise systems, evade safety controls, or facilitate abuse.

Prohibited uses include malware, credential theft, phishing, spam, bot abuse, unauthorized surveillance, doxxing, harassment, extortion, sexual exploitation, child sexual abuse material, terrorist or extremist activity, unlawful weapons activity, evasion of sanctions or export controls, and attempts to bypass security or rate limits.

Customer must not use the Services to generate or assist content that is unlawful, defamatory, infringing, fraudulent, deceptive, or intended to manipulate, impersonate, or mislead people about material facts.

C2. High-Risk and Regulated Uses

Unless expressly authorized in a signed enterprise agreement, Customer must not use the Services for high-risk regulated uses, including credit, lending, insurance, employment, education access, housing, criminal justice, immigration, public benefits, biometric identification, medical diagnosis or treatment, legal advice, financial advice, or other decisions that may produce legal or similarly significant effects for individuals.

Customer must not use the Services as the sole basis for safety-critical, life-critical, emergency, or autonomous decision-making systems.

C3. Security Testing and Abuse Reporting

Customer must not scan, test, attack, overload, scrape, reverse engineer, or interfere with the Services except as expressly authorized in writing.

Security issues, abuse, and suspected misuse should be reported to contact@latens-ai.com.

Appendix D - Service Commitments

D1. Covered Services

These service commitments apply only to dedicated production inference deployments identified as covered in an order form, written confirmation, or signed agreement.

Monthly Availability means the percentage of minutes in a calendar month during which the covered deployment is available to receive inference requests, excluding SLO Exclusions.

D2. Availability SLO

Latens targets 99.95% Monthly Availability for covered dedicated production inference deployments. This is an operational SLO unless a signed agreement states otherwise.

SLO Exclusions include planned maintenance, emergency maintenance, Customer systems or code, Customer network issues, Customer configuration, Customer breach of these Terms, force majeure, denial-of-service attacks, third-party network failures outside Latens reasonable control, suspension for non-payment or abuse, model-specific upstream defects, and periods where Customer is not using the covered deployment in accordance with documented limits.

D3. Service Credits

If a signed agreement, order form, or written confirmation includes service credits and Monthly Availability for a covered deployment falls below the agreed threshold in a calendar month with no SLO Exclusion, Customer may request the applicable service credit. Service credits, where expressly agreed, are the sole and exclusive remedy for failure to meet a service commitment.

To request a service credit where credits have been expressly agreed, Customer must submit a written claim to contact@latens-ai.com within fifteen (15) days after the end of the affected month, including dates, times, logs, request identifiers if available, and a description of the impact.

Service credits, where applicable, are applied to future fees for the affected deployment, are not cash refunds, and are subject to the limits stated in the applicable order form, written confirmation, or signed agreement.

D4. Service Credit Schedule

Service credits apply only where expressly stated in an applicable order form, written confirmation, or signed agreement. If no service-credit schedule is agreed, the 99.95% availability SLO remains an operational target and does not create an automatic credit, refund, or termination right.

D5. Support SLA

For covered production incidents, Latens targets an initial response within six (6) hours when the incident is submitted through the designated support contact. The Support SLA is an initial response commitment only and is not a guarantee of resolution time.

Latens may triage requests based on production impact, severity, security risk, Customer contract terms, and completeness of the information provided. General sales inquiries, onboarding discussions, unpaid evaluations, and non-production requests are not covered production incidents.

Appendix E - Model Terms and Data Handling Summary

E1. Model Information

Latens may make available multiple models with different performance characteristics, context windows, licenses, restrictions, and support status. Customer should review the model information made available for the selected model before production use.

If Customer requires a specific model version, region, license posture, context window, or deprecation notice period, those requirements must be stated in a signed order form or enterprise agreement.

E2. Training, Fine-Tuning, and Evaluation

Latens does not use Customer Content from production inference endpoints to train, fine-tune, evaluate, benchmark, or improve models or services without Customer opt-in or a signed agreement.

Customer-requested fine-tuning, evaluation, benchmarking, prompt logging, or debugging requires separate written terms that define data use, retention, deletion, access, and security controls.