These HUMAIN M3 Limited Preview Model Terms (Model Terms) apply to every person and organization that accesses or uses HUMAIN M3 through Node as part of the Preview, including Guarded Access users and participants in activated Research Access through the Playground or API.
IMPORTANT MODEL NOTICE. HUMAIN M3 is an experimental generative model made available to collect evaluation signals and improve quality, safety and regional or cultural alignment. It is not production-grade. It predicts content rather than retrieving or verifying truth, and safeguards may be incomplete or fail.
OUTPUT RESPONSIBILITY. Outputs may be inaccurate, incomplete, outdated, unsafe, biased, offensive, unlawful, culturally inappropriate or non-unique. Neither HUMAIN nor an underlying model technology provider verifies or assumes responsibility to you for an Output. You are solely responsible for review, use and any consequence of reliance, subject to liability that applicable law does not permit to be excluded.
These Model Terms supplement the Terms of Use, Acceptable Use Policy and current Provider Requirements Register. The Privacy Notice explains data processing. The Research Access Addendum applies only to activated, voluntary Research Access.
You accept these Model Terms through an affirmative electronic control presented with links to these Model Terms and the Provider Register. Acceptance is not consent to personal-data processing; required consents are requested separately through the interface.
1. Model Status and Preview Purpose
The Preview is for evaluation, testing, feedback and non-production prototyping. HUMAIN may use interactions and feedback, subject to the Privacy Notice and required consent, to review, benchmark, train, fine-tune and further develop HUMAIN M3 and its safeguards.
HUMAIN may change the model, weights, system instructions, safeguards, data sources, tools, context limits, behavior, output format and availability at any time. Results may vary across versions, sessions, languages, users and identical prompts.
The Preview does not provide model weights, training datasets, hidden prompts, system configuration or source code and does not commit HUMAIN to any production or open-weight release.
2. Permitted Evaluation Use
Subject to the Preview documents, your enabled access mode and technical controls, you may:
- evaluate Arabic and multilingual capability, dialects, code-switching, reasoning, quality, safety and cultural or regional alignment;
- test controlled, non-production concepts that are not exposed to customers, the public or other unauthorized persons;
- provide ratings, feedback and reproducible reports concerning capability, privacy, security, safeguards or alignment;
- use a guarded API, where enabled, only within the approved organization, use case and technical limits; and
- conduct controlled safety and alignment exploration through Research Access only after voluntary opt-in, verification, acceptance, risk acknowledgement and activation.
Permitted evaluation does not authorize production deployment, public redistribution, third-party service delivery, automated high-impact decision-making or reliance. An Output or technical capability is not authorization.
3. Inputs and Data Restrictions
Submit only Inputs you are lawfully entitled to provide. Do not submit personal, sensitive, confidential, privileged, regulated, classified, export-controlled or security-critical information. Use synthetic or fictitious examples wherever possible.
You are responsible for lawful notices, permissions and consents relating to an Input. HUMAIN may remove, restrict or preserve material and suspend access while investigating suspected unlawful or unauthorized submission.
4. Model Behaviour and Output Limitations
HUMAIN M3 may hallucinate facts, sources, quotations, citations, calculations, translations, code or legal references; misunderstand instructions or context; fail to preserve meaning between languages; and produce inconsistent results.
Arabic limitations may include misunderstanding Modern Standard Arabic, dialects, code-switching, diacritics, transliteration, humor, idioms, religious references, historical context, local law, social norms and cultural nuance. Global use may expose different regional and linguistic gaps.
Safeguards applied by HUMAIN as a filtering layer before and after HUMAIN M3 may under-block harmful content or over-block legitimate content. Research Access may intentionally reduce or remove those safeguards. No mode eliminates the need for independent review and appropriate safety controls.
Outputs may reproduce, resemble or conflict with third-party or training material and may be similar or identical to Outputs supplied to others. HUMAIN does not warrant originality, exclusivity, protectability or non-infringement.
The Model may not know current events and may not accurately identify its source, training data, provider, system settings or limitations. Statements by the Model about HUMAIN, a provider, or these terms are not authoritative.
5. No Reliance or High-Impact Use
Independently verify every Output for accuracy, completeness, legality, appropriateness and fitness for purpose. Apply qualified human judgment before communicating, publishing, implementing or acting on an Output.
Do not use an Output as legal, medical, financial, investment, tax, religious, employment, educational, governmental, emergency or other professional advice, or as the sole or material basis for a decision affecting a person's rights, safety, health, livelihood, reputation, finances or access to an essential service.
Do not use HUMAIN M3 in nuclear facilities, air-traffic control, life-support, critical infrastructure or another environment where use or failure could reasonably cause death, injury, significant property damage or environmental harm.
Do not follow, execute, share or operationalize harmful instructions. Do not run code, open links, contact targets, access systems, acquire materials or take real-world action based on an unsafe Output.
6. Research Access
Any eligible adult user may voluntarily request Research Access. HUMAIN may select, verify, approve, condition, pause or revoke access. No academic qualification, professional affiliation, research history or formal test plan is required unless HUMAIN requests specific information for verification or risk controls.
Research Access begins only after HUMAIN provides an activation notice identifying the enabled access period, model or safeguard mode and applicable limits. It is personal, confidential, available through the Playground and, where enabled, the API, and subject to the Research Access Addendum.
HUMAIN may review every Research Access interaction and use Inputs, Outputs, ratings and feedback for safety review, evaluation, training, fine-tuning and further development subject to the Privacy Notice and required electronic consents.
Reduced safeguards do not create permission to cause harm, target real people or systems, possess or distribute unlawful material, operationalize harmful content or publish Research Materials.
7. Recording, Review and Model Improvement
Every Input and Output is recorded with related account and technical data. Authorized HUMAIN and provider reviewers may analyze any interaction under confidentiality, access-control and data-protection obligations. The model technology provider may receive and use Preview data for inference, review, safety, evaluation, training and further development of HUMAIN M3 and its related safeguards, in accordance with HUMAIN's provider arrangements and applicable law.
Raw, user-linked interactions will ordinarily be retained for twelve (12) months. Selected effectively de-identified data may be retained longer for safety, evaluation, training and development where re-identification is not reasonably possible.
Separate affirmative electronic consent is required for the interaction processing described in the Preview interface and Privacy Notice. Withdrawal affects future consent-based processing and ends new interactions, without affecting processing already lawfully completed.
8. Model and Output Rights
HUMAIN and its licensors retain all rights in HUMAIN M3, Node, the Preview, software, interfaces, documentation, safeguards and underlying technology. No right to weights, source code, training data, hidden prompts or system configurations is granted.
To the extent permitted by law and as between you and HUMAIN, you may use an Output generated for you subject to the Preview documents. You are responsible for clearing third-party rights and for any disclosure, publication or implementation. Research Materials remain subject to confidentiality and no-publication obligations.
You must not reverse engineer, decompile, disassemble, extract or attempt to reconstruct the Model or protected systems; systematically harvest Outputs; evade controls; or use the Preview or Outputs to train, benchmark or support a competing model, dataset, product or service.
9. Provider Requirements and Changes
The current Provider Requirements Register identifies downstream-relevant model restrictions. It supplements these Model Terms where silent and the more restrictive applicable rule controls within its stated scope. It does not create a direct provider-user contract.
HUMAIN may change or withdraw the Model or update these Model Terms and the Register. Material contractual changes require renewed affirmative electronic acceptance before further use, and processing changes require fresh consent where applicable.
10. Model Disclaimers and Liability
TO THE MAXIMUM EXTENT PERMITTED BY LAW, HUMAIN M3, THE PREVIEW AND OUTPUTS ARE PROVIDED 'AS IS' AND 'AS AVAILABLE' WITHOUT WARRANTIES OF ACCURACY, COMPLETENESS, SAFETY, LEGALITY, NON-INFRINGEMENT, MERCHANTABILITY, FITNESS FOR PURPOSE OR AVAILABILITY.
TO THE MAXIMUM EXTENT PERMITTED BY LAW, NEITHER HUMAIN NOR AN UNDERLYING MODEL TECHNOLOGY PROVIDER IS RESPONSIBLE OR LIABLE TO YOU FOR AN OUTPUT, A SAFEGUARD FAILURE, A MODEL CHANGE OR A DECISION, COMMUNICATION, PUBLICATION, IMPLEMENTATION OR ACTION BASED ON AN OUTPUT. THE LIABILITY LIMITS IN THE TERMS APPLY.
Nothing excludes liability, mandatory rights or remedies that applicable law does not permit to be excluded. References to the underlying provider create no direct contract, benefit or enforcement right.
11. General
The governing law, disputes, language, notices, assignment, severability and contact provisions in the Terms apply to these Model Terms. Questions and reports may be submitted through the relevant channel identified in the Preview interface or Privacy Notice.