1. Agreement and Definitions#

1.1 Who we are. DATUMO (the "Service") is provided by Datumo, the name under which 셀렉트스타 주식회사, a company incorporated in the Republic of Korea, operates outside Korea, with its office at 8th–11th floors, 8 Teheran-ro 38-gil, Gangnam-gu, Seoul 06221, Republic of Korea ("Datumo," "we," "us" or "our").

1.2 Agreement. These Terms of Service (the "Terms") govern your access to and use of the Service, including the DATUMO Console, Quick Eval, our websites, APIs, SDKs, documentation and related services. You accept these Terms when you create an account, click to accept them, upload content to Quick Eval, accept an invitation, or otherwise use the Service. If you do not agree, do not use the Service.

1.3 Using the Service for an organization. If you use the Service on behalf of a company or other entity, you confirm that you are authorized to accept these Terms for it. In that case "you" and "Customer" mean that entity, and the entity is responsible for its Users' use of the Service.

1.4 Eligibility. The Service is intended for business and professional use. You must be at least 18 years old, or the age of majority where you live if higher, and able to form a binding contract. You may not use the Service if applicable law, including sanctions law, prohibits you from doing so.

1.5 Related documents. Any order form or enterprise agreement you sign with us (an "Order Form") forms part of these Terms. If an Order Form conflicts with these Terms, the Order Form prevails. Our Privacy Policy explains how we handle personal information as a controller; it is not part of these Terms.

1.6 Definitions.

In these Terms, the "Member role" means the Organization role described in Section 3.5. "Customer" means the company or other entity for which an Organization is used or, if there is none, the individual who owns it. A "Fail Case" is a result that fails a metric or threshold you set.

2. The Service#

2.1 Overview. DATUMO is a platform for evaluating, testing and monitoring AI systems. You can design evaluation criteria, run automated and human evaluations, simulate users, red-team AI systems, monitor production logs and analyze results. The features available to you depend on phased releases (Section 2.4).

2.2 Features. The Service includes the following features:

FeatureWhat it does
ConsoleSign-up and sign-in; Organization, Workspace, Member, role and invitation management; model, Application and integration settings
Quick EvalEvaluates 50–500 uploaded conversation logs: AI extracts evaluation tasks, drafts policies, generates metrics, scores the logs and suggests improvements. It can be used before sign-up, with limited results
CriteriaBuilds evaluation frameworks (Task Groups, Tasks, Metrics, Benchmark Sets and Release Gate thresholds) using LLM-as-a-Judge, rule-based, embedding, decision-model and Python-code metrics. AI Tools generate, modify, augment and validate benchmark data
RunsExecutes evaluations: the Service sends each benchmark row to the selected Application versions, collects the responses and scores them with judge models
Evaluation DashboardRun reports with Pass/Fail verdicts, version comparisons and failure clusters; custom chart pages, including an AI chart builder; usage and cost analytics; report links
Agent SimulationGenerates scenarios and synthetic personas from your agent policy and tool schema. An AI-simulated user converses with your agent, and the conversations and tool-call trajectories are scored
Red TeamingAutomated adversarial testing: attacker models generate attack prompts from seed sets and algorithms and send them to your Application, and judge models score the responses. Produces attack success rates, heatmaps, exports and reports
Evaluation Lab and Annotation QueueHuman red-teaming and review tasks: your testers interact with your Application, and submissions are judged by the testers or by AI rubrics
ObservabilityConnects production logs from your cloud account (currently AWS CloudWatch), evaluates sampled conversation turns against metrics, records Fail Cases and sends alerts
Gen & Judge ModelsRegisters and manages the models used for generation, simulation, attacks and judging

2.3 AI notice. The Service is operated using artificial intelligence, including generative AI. Many Outputs are generated by AI models, including scores, judge reasons, generated criteria, policies, metrics, benchmark data, scenarios, personas, attack prompts, chart settings and improvement suggestions. The Service shows which models are used, and we identify AI-generated Outputs as required by law, including Korea's Framework Act on the Development of Artificial Intelligence and Establishment of Trust (the "AI Basic Act").

2.4 Phased availability. Some features may be released in phases or offered to selected accounts first. For example, some features, such as Agent Simulation and Red Teaming, may not be available to run at launch. The Service shows which features are available to you. Features labeled "Beta," "Preview," "Coming soon" or similar are Beta Features (Section 8.1).

2.5 Changes to the Service. We may add, change or remove features, models, Datumo Materials and Usage Limits to improve the Service, meet legal or security requirements, or reflect changes by model providers. Where practicable, we will notify you in advance of changes that materially reduce the core functionality of the Service.

3. Accounts, Organizations and Members#

3.1 Your account. You can sign up with a work email address, which we verify by sending a link, or with a supported third-party sign-in (currently Google or Microsoft). You must give accurate information and keep it current. If you sign in with Google or Microsoft using an email address that matches an existing DATUMO account, the two are linked, and you can use either method to sign in.

3.2 Account security. You are responsible for keeping your password, sign-in methods, DATUMO API keys and other credentials secure, and for all activity under your account. Tell us promptly at [email protected] if you suspect unauthorized use. We may require re-authentication, such as entering your password, before sensitive actions such as transferring ownership or deleting an Organization or account.

3.3 Organizations. When you sign up other than by accepting an invitation, you create an Organization, become its Owner and name its first Workspace. Each account may own only one Organization that it creates itself; you may join other Organizations by invitation and switch between them. If an Organization is created for a company or other entity, that entity is the Customer, and the Owner accepts these Terms on its behalf.

3.4 Workspaces and Applications. Owners and Admins may create and delete Workspaces within the Usage Limits shown in the Service; at launch, each Organization is limited to one Workspace. Owners, Admins and Workspace Admins may create and delete Applications in their Workspaces. Deleting a Workspace deletes its Applications, runs, results, datasets, metrics and model settings as described in Section 4.9, and its Members lose access to it.

3.5 Roles and permissions. Access is controlled by roles at two levels:

RoleLevelMain permissions
OwnerOrganization (exactly one)Everything an Admin can do, plus transferring ownership and deleting the Organization
AdminOrganizationInvite and remove Members; assign roles, including Admin; create and delete Workspaces
Member (role)OrganizationAccess only the Workspaces assigned to them, with the Workspace role given
Workspace AdminWorkspaceManage Workspace settings, Workspace roles, models and Applications; create and run evaluations
EditorWorkspaceCreate, edit and run evaluations outside Settings; runs count toward Usage Limits
ViewerWorkspaceView and export results and dashboards

Owners and Admins automatically have Workspace Admin rights in every Workspace. Some settings, such as Organization-level model registration, are limited to the Owner. We may add or adjust roles and permissions, and custom roles may be offered under an Order Form.

3.6 Invitations. Owners and Admins may invite people by email and set their Organization role and Workspace access. Invitations expire after 7 days. An invitation can be resent, which invalidates the earlier link, or revoked. Pending invitations count toward any member limit that applies to your Organization. An invitation can be accepted only with the invited email address, and invitees who create an account must accept these Terms. The invitee will see the inviter's name and email address and the Organization's name. By sending an invitation, you confirm that you may share the invitee's email address with us for this purpose.

3.7 Responsibilities of Owners and Admins. Owners and Admins control access to the Organization and its Customer Content. They are responsible for: (a) granting and removing access promptly, including for people who leave; (b) the models, Customer Systems, credentials and alert recipients configured for the Organization; and (c) making sure Members comply with these Terms. Members acknowledge that Owners and Admins, and the Customer, can view and manage the content Members create in the Organization.

3.8 Removing Members and leaving. Owners and Admins may remove a Member from the Organization or from a Workspace at any time. The Member loses access immediately, and the content they created stays in the Organization. A Member may leave an Organization at any time, but the Owner must transfer ownership before leaving. A re-invited User joins with the role in the new invitation; earlier roles are not restored.

3.9 Transferring ownership. The Owner may transfer ownership to an Admin who does not already own an Organization. The former Owner becomes an Admin. If the Owner is no longer available, for example because they have left the Customer, we may transfer ownership at the Customer's verified request.

3.10 Deleting an Organization. In the Service, only the Owner may delete an Organization; we may also delete one at the Customer's verified request or under Section 11. Deletion removes all its Workspaces, Applications, Customer Content and settings, ends all Members' access, cancels pending invitations and stops Observability log collection. Jobs that are already running may continue until they finish, and their results are deleted with the Organization. We keep a deleted Organization for 30 days, during which the Owner may ask our support team to restore it; after that it is permanently deleted as described in the Privacy Policy. An Organization deleted together with its Owner's account cannot be restored at the Owner's request (Section 3.11).

3.11 Deleting your account. You may delete your account in My Account. If you are the only Member of an Organization you own, that Organization is deleted with your account and cannot be restored at your request. If an Organization you own has other Members, you must first choose a new Owner. When you delete your account, you leave all Organizations, your DATUMO API keys are revoked, invitations you sent remain until they expire, and content you created stays in each Organization under its control. You can sign up again with the same email address after 14 days. Once your account is deleted, you can no longer ask us to restore Organizations you owned.

3.12 Access by Datumo personnel. Authorized Datumo personnel can access every Organization through internal administration tools, with the ability to view content and to change settings, roles, ownership and deletion status. We use this access only to provide support you request, maintain and secure the Service, investigate suspected violations of these Terms and comply with law. We change roles, ownership or deletion status only at the Customer's verified request or under Section 11. Access is limited to personnel who need it, is logged and is subject to confidentiality obligations.

3.13 Enterprise and custom deployments. Accounts that Datumo provisions for enterprise, on-premises or customer-specific deployments, including administrator-provisioned accounts with custom password or two-factor authentication policies, are governed by the applicable Order Form. These Terms apply where the Order Form does not say otherwise.

4. Customer Content#

4.1 Ownership. As between you and Datumo, you own your Customer Content, including Inputs and, to the extent the law allows, Outputs. We do not claim ownership of Outputs generated for you. Because of how AI works, Outputs may not be unique, and other customers may receive similar Outputs. Outputs do not include Datumo Materials that appear in them, which remain subject to Section 9.3.

4.2 License to Datumo. You grant Datumo a worldwide, non-exclusive, royalty-free license to host, copy, process, transmit, display and modify Customer Content only as needed to: (a) provide, maintain and secure the Service for you; (b) prevent fraud, abuse and security incidents; (c) provide support you request; and (d) comply with law. This includes sending Customer Content to your Customer Systems, to the models selected in your settings, and to our subprocessors listed in the Privacy Policy. The license lasts while we process Customer Content under these Terms.

4.3 No training on Customer Content. We do not use Customer Content to train, fine-tune or otherwise improve AI models, including Datumo Models, or to build datasets or benchmarks for others, unless you expressly opt in through a clearly labeled setting or in writing. We may use aggregated and de-identified information about use of the Service, such as feature usage and performance statistics that do not identify you, your Users or your end users and do not reveal Customer Content, to operate and improve the Service.

4.4 Your responsibilities. You represent and warrant that: (a) you have all rights and permissions needed to submit Customer Content and to let us process it as described in these Terms; (b) you have given all notices to, and obtained all consents from, individuals as required by law, including your end users, employees and testers, for example before collecting their conversation logs and having them analyzed by third-party service providers; (c) your Customer Content, and our processing of it on your instructions, will not violate law or anyone's rights; and (d) you comply with the terms of any third-party service you connect.

4.5 Personal and sensitive data. Submit only the personal data you need. Where feasible, mask, pseudonymize or synthesize personal data before submitting it; Quick Eval and AI-generated criteria expect masked logs. Do not submit the following unless it is strictly necessary for your evaluation, you have a lawful basis and any required consents, and you apply appropriate safeguards:

Never submit unmasked resident registration numbers, passport numbers, social security numbers or similar national identifiers. We do not check whether Customer Content contains such data, and you are responsible for any you submit.

4.6 Our role for personal data in Customer Content. For personal data in Customer Content, the Customer is the controller, and Datumo processes the data on the Customer's behalf as a processor (a trustee under Korea's Personal Information Protection Act and a service provider under US state privacy laws). We process it only on your documented instructions, which include these Terms and your configuration of the Service, and under Section 10.2.

4.7 Outputs. Outputs are produced by AI and automated methods and may be inaccurate, incomplete, biased or offensive. Review Outputs before relying on them; Section 12.1 explains the limits of evaluation results. Synthetic personas and scenarios generated by the Service are fictional. Any resemblance to real people is coincidental, and you must not present them as real.

4.8 Sharing and exports. Users can export results and copy links to reports, sessions and Fail Cases, according to their role. Exports and links may contain Customer Content, including raw inputs, model responses, judge reasons and any personal data they include. Share them only with people authorized to see that content. You are responsible for exported files once they leave the Service.

4.9 Deletion and changes to results. You can delete Customer Content in the Service, for example runs, datasets, simulations, dashboard pages, Workspaces and Organizations. Some items cannot be deleted while processing is in progress, and many deletions cannot be undone. Items you delete are removed from the Service immediately but may remain in our systems until the Organization is deleted, unless you ask us at [email protected] to delete them sooner. Some configuration changes apply to existing results: renaming a key or metric, moving a Task or changing a threshold may change how past results are labeled, grouped or judged. Export results you need to keep in their current form.

5. Models, Customer Systems and Integrations#

5.1 Model types. The Service uses three kinds of models and shows which one is used for each role, such as generation, simulated user, attacker, judge or embedding:

5.2 Provided Third-Party Models. When you use a Provided Third-Party Model, we send the relevant Inputs to its provider, which acts as our subprocessor. You must comply with that provider's usage policies, such as the OpenAI Usage Policies, the Anthropic Usage Policy and Google's Generative AI Prohibited Use Policy. We may restrict or disable models, content categories or features to comply with provider policies or law.

5.3 Customer-Registered Models. When you register a model with your own credentials, we send data to that provider or endpoint on your instructions. That provider is your vendor, not our subprocessor, and your agreement with it governs its fees, data use, retention and usage policies. The provider bills you directly for that usage. We are not responsible for the availability, security, outputs or conduct of Customer-Registered Models.

5.4 Customer Systems. When you connect an Application, agent, model endpoint or log source, you authorize us to access it and send requests to it as needed for the features you use. Examples include benchmark inputs in Runs, simulated user turns in Agent Simulation, attack prompts in Red Teaming and tester messages in Evaluation Lab. You are responsible for: (a) your right to connect the system and to authorize this traffic; (b) the volume of requests, which may be large (hundreds or thousands per run), and any resulting costs, rate limits and load; (c) securing the system and the credentials you give us; and (d) everything the system does in response.

5.5 Agents and live systems. Agent Simulation, Red Teaming and Evaluation Lab can cause your agent to call tools or take actions, such as sending messages, changing records or initiating payments. Use test or staging environments, test accounts and sandboxed tools whenever possible, and connect a production system only if you accept the consequences. Datumo does not control actions your systems take and is not responsible for them, except to the extent caused by our breach of these Terms, gross negligence or intentional misconduct.

5.6 Credentials. You may give us credentials, such as model API keys, request headers containing tokens, and AWS access keys. We store them encrypted, mask values you mark as secret, and use them only to provide the Service. Grant only the minimum permissions needed, for example read-only access to the specific log groups Observability uses. Rotate credentials regularly and revoke them when you no longer need them. You are responsible for activity performed with credentials you provide.

5.7 DATUMO API keys. Where the Service offers DATUMO API keys, each key belongs to the User who creates it and is scoped to one Organization. Owners and Admins can view and revoke all keys in their Organization, and a key never has more permissions than its User's role. Do not share keys outside your Organization. Activity performed with a key is attributed to its User and Organization. Keys are revoked when their User leaves the Organization or deletes their account.

5.8 Code you provide. Some features run code you provide, such as Python Code metrics and Observability parser code. You are responsible for that code. It must not contain malware, access systems or data beyond what it is meant to process, circumvent isolation, security or Usage Limits, or make network calls it does not need. We may limit execution time, memory, libraries and network access, and may stop or block code that poses a risk.

5.9 Observability. When you connect a log source (currently AWS CloudWatch), you authorize us to retrieve logs from the log path you specify on an ongoing basis, and to run your parser code on them, until you disable the integration in the Service, ask us to disable it at [email protected], delete the Application, or the Organization is deleted. You decide which logs are collected. Send only what you need for evaluation, and filter out unrelated data such as system logs, IP addresses and identifiers. Evaluation is sample-based according to your sampling settings, and turns that were not evaluated will not be evaluated later. Fail Case alert emails include the end user's input and the Application's output for the failed turn, and you choose their recipients.

5.10 Third-party services. The Service works with third-party services such as Google and Microsoft sign-in and AWS. Their own terms govern your use of them, and we are not responsible for them.

6. Acceptable Use#

6.1 General rules. You must not, and must not allow anyone to:

6.2 Test only what you are authorized to test. You may use Runs, Agent Simulation, Red Teaming, Evaluation Lab and Observability only on Customer Systems that you own or operate, or that you are expressly authorized to test and to send this traffic to. If you test a third-party model or service, including one accessed through a third party's API, you must follow that third party's terms, including its rules on safety testing, rate limits and prohibited content.

6.3 Adversarial and harmful content. Red Teaming and Evaluation Lab are designed to generate adversarial prompts and to capture harmful or offensive responses for safety testing. You may use this content only to test, evaluate, document and improve the safety of AI systems. You must not:

We may block, remove or preserve content that violates this Section, and report it to competent authorities or organizations where the law requires.

6.4 Safety controls. To comply with law and provider policies, we may apply automated filters, category restrictions, masking, rate limits or human review to adversarial features. We may also stop runs or restrict features that we reasonably believe pose legal or safety risks.

6.5 Human testers and reviewers. If you assign people to Evaluation Lab tasks, the Annotation Queue or result review, you must tell them they may see harmful, offensive or disturbing content; obtain any consents the law requires; provide appropriate safeguards and support; and ensure they are adults authorized to do the work. Tester statistics, such as attack success rates per tester, are Customer Content, and you are responsible for how you use them in relation to your personnel.

6.6 Simulations and synthetic personas. Scenarios and personas are generated from your policy and are fictional. Do not include real individuals' personal data in policies, persona descriptions or task instructions unless it is necessary and lawful. Section 5.5 applies to actions your agent takes during simulations.

6.7 Export controls and sanctions. You must comply with the export control and sanctions laws of Korea, the United States, the European Union, the United Kingdom and other applicable countries. You represent that you are not located in, or owned or controlled from, a country or region subject to comprehensive sanctions, and that you are not on a restricted-party list. You must not use the Service for prohibited end uses, including the development of weapons.

6.8 Enforcement. We may investigate suspected violations of this Section and suspend or restrict access as described in Section 11.3.

7. Usage Limits#

7.1 Free of charge. The Service is currently provided free of charge. Before we charge for any part of the Service, we will amend these Terms under Section 13.3.

7.2 Usage Limits. Your use of the Service is subject to the Usage Limits shown in the Service. When you reach a limit, some actions may be blocked or stopped until the limit resets or is changed. Some runs cannot be paused or cancelled once started, and a run may stop partway if a limit is reached.

7.3 Displayed costs. Cost, token and latency figures in the Service are estimates based on recorded usage and on prices that you or we enter. They are not invoices. Providers of Customer-Registered Models and of your Customer Systems bill you directly for their charges.

8. Beta Features, Quick Eval, Availability and Support#

8.1 Beta Features. Beta Features are offered for evaluation, may be incomplete, and may change or be discontinued at any time. They are excluded from any service level commitment and, to the extent the law allows, are provided without warranties. Customer Content in Beta Features is handled like other Customer Content unless we tell you otherwise.

8.2 Quick Eval before sign-up. You can use Quick Eval without an account. By uploading logs or a policy, you accept these Terms and confirm that you have the right to upload them and that personal information in them has been masked. You upload Quick Eval content as, or on behalf of, the controller of any personal information in it, such as the organization you work for. We process that content as a processor under Sections 4.6 and 10.2, only to produce your Quick Eval results, and we do not use it to train or improve AI models or for any other purpose. Quick Eval content is Customer Content for the purposes of Sections 4 and 9.6. Guest results are linked to your browser session and carry over to an account only if you sign up for a new account after the results appear. If you sign in to an existing account instead, the session's results do not carry over. Guest uploads that do not carry over are kept for 7 days and then deleted. Sample scenarios use pre-computed results and synthetic data.

8.3 Availability. We aim to keep the Service available at all times, except for planned and emergency maintenance and events beyond our control. We will announce planned maintenance in advance where practicable. Unless an Order Form includes a service level agreement, we do not guarantee any level of availability.

8.4 Support. Support is available through the channels shown in the Service and documentation, including documentation and email support. You can contact us at [email protected].

8.5 Communications. We send service messages about your account, security, invitations, usage and changes to these Terms by email or in the Service. These are part of the Service, and you cannot opt out of them while you have an account. We send marketing messages only with your consent, and you can unsubscribe at any time.

9. Intellectual Property, Feedback and Confidentiality#

9.1 Our rights. Datumo and its licensors own the Service, Datumo Models, Datumo Materials, software, documentation and trademarks, including DATUMO. These Terms grant you only the rights they expressly state.

9.2 Your right to use the Service. Subject to these Terms and any Order Form, we grant you a limited, non-exclusive, non-transferable and non-sublicensable right to access and use the Service while these Terms are in effect, and to let your Users do so, for your internal business purpose of evaluating, testing and monitoring AI systems.

9.3 Datumo Materials. You may use Datumo Materials only within the Service and for your internal purposes, and you may export results that include them for internal use. Without our written consent, you may not extract or copy Datumo Materials in bulk; publish, sell or redistribute them, for example as a dataset, benchmark or attack library; or use them to train AI models. Some Datumo Materials include or are based on open-source software, datasets or methods, which we identify in the documentation, and their own licenses govern those components.

9.4 Publishing results. If you publish results produced with the Service, such as model comparisons, describe your methodology accurately and do not state or imply that Datumo endorses or certifies the results. You may say that you used DATUMO.

9.5 Feedback. If you send us feedback or suggestions, we may use them without restriction or obligation to you.

9.6 Confidentiality. Each party may receive non-public information from the other ("Confidential Information"), and your Customer Content is your Confidential Information. The receiving party will use Confidential Information only to perform under these Terms, protect it with at least reasonable care, and disclose it only to employees, contractors and subprocessors who need to know it and are bound by similar duties. These duties do not cover information that is or becomes public through no fault of the receiving party, was already known to it, is independently developed, or is rightfully received from a third party. The receiving party may disclose Confidential Information when the law requires, after giving notice where the law allows. These duties continue for 3 years after these Terms end, and for trade secrets for as long as they remain trade secrets.

9.7 Publicity. We will not use your name or logo in marketing without your prior consent.

10. Data Protection and Security#

10.1 Privacy Policy. Our Privacy Policy explains how we handle the personal information of Users, invitees and visitors, for which Datumo is the controller.

10.2 Processing terms. When we process personal data in Customer Content as your processor, we will: (a) process it only on your documented instructions, including these Terms and your settings, unless the law requires otherwise, and tell you if we believe an instruction breaks the law; (b) not use it for any other purpose, not sell or share it, and not retain, use or disclose it outside our direct business relationship with you or combine it with personal information from other sources, except as the law permits; (c) make sure the personnel who process it are bound by confidentiality; (d) protect it with the measures in Section 10.5; (e) use subprocessors only as described in Section 10.3, under written terms no less protective than this Section, and remain responsible for them; (f) transfer it to other countries only as described in the Privacy Policy and with the safeguards the law requires; (g) help you, taking into account the nature of the processing, to respond to requests from individuals and to meet your obligations on security, breach notification and data protection impact assessments; (h) notify you without undue delay after becoming aware of a personal data breach affecting it; (i) delete it as described in Sections 3.10, 4.9 and 11.6 and the Privacy Policy, unless the law requires us to keep it; and (j) make available the information reasonably needed to demonstrate compliance with this Section, first by providing, under confidentiality, our most recent SOC 2 Type II report and answers to reasonable security questionnaires, and, only where that information is not sufficient, cooperate with audits by you or an auditor you appoint, on reasonable notice, no more than once a year unless a supervisory authority requires it or after a breach, and at your expense. This Section contains the terms required by Article 28 of the GDPR and the UK GDPR, Article 26 of Korea's Personal Information Protection Act, and the California Consumer Privacy Act and other US state privacy laws. If you need a separately signed data processing agreement, contact [email protected].

10.3 Subprocessors. You authorize us to use subprocessors, such as cloud hosting, GPU hosting, AI model and email providers. We list them in the Privacy Policy and update that list at least 30 days before we add or replace a subprocessor that processes Customer Content. You may object on reasonable data protection grounds by emailing [email protected] within that period. If we cannot resolve your objection, you may stop using the affected feature or delete your Organization.

10.4 Data location. We host the Service and store Customer Content mainly in the Amazon Web Services Seoul region in the Republic of Korea. Customer Content may be processed in other countries when it is sent to models, Customer Systems or subprocessors, as described in the Privacy Policy.

10.5 Security. We maintain administrative, technical and physical safeguards appropriate to the risk, including encryption in transit and at rest, least-privilege access controls, masking of credentials you mark as secret, logging and monitoring, and staff confidentiality and training. Our security controls are independently audited under SOC 2 Type II. If we become aware of a security incident affecting Customer Content, we will notify affected Customers without undue delay and as the law requires, and give them the information they reasonably need to meet their own obligations.

10.6 Audit logs. We record certain activity, such as role changes, invitations, deletions and administrative access, for security and compliance.

10.7 Your security responsibilities. You are responsible for configuring roles, sharing, alert recipients and credentials appropriately; for securing your devices and Customer Systems; and for exporting or backing up results you need to keep.

11. Term, Suspension and Termination#

11.1 Term. These Terms apply from when you first accept them or use the Service until your account is deleted or these Terms are terminated. An Order Form may set its own term.

11.2 Ending use. You may stop using the Service at any time. You may delete Workspaces, delete an Organization you own (Section 3.10) or delete your account (Section 3.11). You may also terminate these Terms by notice if we materially breach them and do not cure the breach within 15 days after your notice.

11.3 Suspension. We may suspend or limit all or part of the Service, including specific runs, integrations or features, if: (a) you materially breach these Terms, including Section 6; (b) you materially breach an Order Form; (c) your use poses a security risk or could harm the Service, other customers or third parties; (d) the law, a court order or a competent authority requires it; or (e) we reasonably suspect fraud or abuse, such as creating multiple accounts to avoid Usage Limits. We will give notice in advance where practicable, and we will restore access once the issue is resolved.

11.4 Termination by us. We may terminate these Terms or your access: (a) if you materially breach these Terms and do not cure the breach within 15 days after notice; (b) immediately, for serious violations such as illegal content, child sexual abuse material, unauthorized testing of third-party systems or attacks on the Service; or (c) if we discontinue the Service, with at least 60 days' notice.

11.5 Effect of termination. When these Terms end, your right to use the Service ends. We delete Customer Content as described in Sections 3.10 and 3.11 and in the Privacy Policy. Export any Customer Content you want to keep before you delete it or before your access ends.

11.6 Data export and switching. When these Terms end, other than because you deleted the content, Workspace or Organization yourself, you may export your Customer Content with the Service's export features for at least 30 days afterwards, unless we terminated under Section 11.4(b) and the law allows us to refuse. After that period, we delete it as described in the Privacy Policy. On request, we will provide reasonable assistance to export your Customer Content in a common machine-readable format, such as CSV or JSON, to help you switch to another provider, as required by applicable law, including the EU Data Act where it applies.

11.7 Survival. Sections 4.4, 4.8, 6.3, 9, 11.5, 11.6, 12 and 13, Section 10.2 for as long as we hold Customer Content, and any other terms that by their nature should survive, survive the end of these Terms.

12. Disclaimers, Limitation of Liability and Indemnity#

12.1 Evaluation results are not guarantees. The Service helps you assess AI systems, but its results come from automated methods and AI models that can be wrong. This applies to scores, verdicts such as Pass/Fail and Release Gate decisions, attack success rates, Fail Cases and recommendations. Automated judges can produce false positives and false negatives; benchmarks, samples and simulations may not reflect real-world use; and Observability covers only the logs and samples you configure, so it may miss issues or alert late. Results are not a certification, audit or guarantee that an AI system is safe, accurate, fair, secure or lawful, and they are not legal, medical, financial or other professional advice. You remain solely responsible for decisions to develop, release and operate your AI systems and for complying with the laws that apply to them, such as the EU AI Act and Korea's AI Basic Act.

12.2 Disclaimer. Except as expressly stated in these Terms or an Order Form, and to the extent the law allows, the Service, Outputs, Datumo Models, Datumo Materials and Beta Features are provided "as is" and "as available," without warranties of any kind, whether express, implied or statutory, including warranties of merchantability, fitness for a particular purpose, accuracy, non-infringement, and uninterrupted or error-free operation.

12.3 Limitation of liability. To the extent the law allows: (a) neither party is liable for indirect, incidental, special, consequential or punitive damages, or for lost profits, revenue, goodwill or data, even if told they were possible; and (b) each party's total liability arising out of or relating to these Terms is limited to USD 100, unless an Order Form sets a different amount. These limits do not apply to: (i) liability for intentional misconduct or gross negligence; (ii) your obligations under Section 12.4 and your breaches of Section 6; or (iii) liability that cannot be limited under applicable law, including mandatory consumer protection law.

12.4 Your indemnity. You will defend and indemnify Datumo and its affiliates, directors and employees against third-party claims, and related losses, damages, fines and reasonable costs, to the extent they arise from: (a) your Customer Content, including claims that it infringes rights or was collected or processed without the notices or consents the law requires; (b) your Customer Systems and their actions; (c) your use of the Service in breach of these Terms, including Section 6; or (d) your breach of law or of third-party terms, including the usage policies of providers of Customer-Registered Models. We will notify you promptly of a claim, let you control its defense and cooperate reasonably at your expense. You may not settle a claim in a way that imposes obligations on us without our consent. Any additional indemnities are set out in an Order Form. If you are a consumer, this Section applies only to the extent the law allows.

12.5 Force majeure. Neither party is liable for failure or delay caused by events beyond its reasonable control, such as natural disasters, war, terrorism, epidemics, labor disputes, government action, or failures of networks, cloud providers or model providers.

13. Governing Law, Changes and General Terms#

13.1 Governing law. These Terms are governed by the laws of the Republic of Korea, without regard to conflict-of-laws rules. The UN Convention on Contracts for the International Sale of Goods does not apply.

13.2 Disputes. The parties will first try to resolve any dispute through good-faith negotiation for 30 days. Unresolved disputes are subject to the exclusive jurisdiction of the Seoul Central District Court, unless an Order Form provides otherwise. If you are a consumer, you may also bring proceedings in the courts of your place of residence where mandatory law gives you that right. Consumers in Korea may also apply for mediation to the Korea Consumer Agency or the Electronic Commerce Mediation Committee.

13.3 Changes to these Terms. We may amend these Terms as needed, in compliance with applicable law. We will post the amended Terms on our website and in the Service with their effective date and a summary of the changes, and notify all Users by email or in-product notice, and Owners and Admins by email. We will do so at least 7 days before the effective date, or at least 30 days before for changes that are unfavorable to you or materially affect your rights. The notice will state clearly that continuing to use the Service after the effective date means you accept the amended Terms. If you do not agree, you may stop using the Service and delete your account before the effective date. Where the law requires your express consent, we will ask for it.

13.4 Transition from the previous Terms. These Terms replace the DATUMO Terms and Conditions of Service dated January 14, 2026. We notify Users who accepted the previous Terms in advance, and we ask them to accept these Terms when they next sign in on or after November 11, 2026.

13.5 Assignment. You may not assign these Terms without our prior written consent. We may assign them, with notice, in connection with a merger, acquisition, reorganization or sale of assets.

13.6 Notices. We send notices to the email address of the relevant User, Owner or Admin, or in the Service. You can send legal notices to [email protected].

13.7 General. These Terms, together with the documents in Section 1.5, are the entire agreement between you and Datumo on their subject. If any provision is unenforceable, the rest remain in effect. A failure to enforce a provision is not a waiver. The parties are independent contractors. These Terms are available in English and Korean. For customers in Korea, the Korean version prevails; for all others, the English version prevails.

13.8 Contact. Datumo (셀렉트스타 주식회사), 8th–11th floors, 8 Teheran-ro 38-gil, Gangnam-gu, Seoul 06221, Republic of Korea. Support and legal notices: [email protected]. Privacy: [email protected].