1. Additional conditions
These conditions supplement the operome.dev Terms of Use. Our Privacy Policy explains how personal data is handled. By submitting a URL, confirming facts, requesting a classification, unlocking a result or using a permalink, you accept these conditions and the Terms of Use.
2. Informational tool only
The classifier is a research and demonstration tool. A result is not legal advice, a legal opinion, a regulatory determination, a conformity assessment, certification, assurance report, audit, approval or SynapseLayer attestation. Regulators, courts, notified bodies and other authorities are not bound by it.
3. Limited scope
The classifier addresses only the rules and corpus version identified in the result, presently including selected questions about EU scope, prohibited practices, Article 6 high-risk routes and general-purpose AI model classification. It does not determine all obligations, exceptions, transition dates or sector-specific rules.
A result does not replace assessment of transparency duties, provider and deployer obligations, quality management, technical documentation, fundamental-rights impact assessment, data protection, cybersecurity, product safety, employment, consumer, intellectual-property or other applicable law.
4. Website extraction is provisional
Automated extraction uses a managed third-party AI model to analyse only the public pages that can be retrieved at the time. The provider and model used for a scan are disclosed in the review screen and may change—for example, the service may use an OpenAI, Anthropic/Claude, Google/Gemini or another configured model. The submitted URL and retrieved public-page text may be transmitted to that provider. Do not submit confidential, personal or sensitive information.
Extraction may miss dynamic, blocked, ambiguous or unpublished information and may infer a fact incorrectly. Extracted answers are proposals for human confirmation, not verified facts. “Unknown” is intentional and must not be treated as “No”. Generative AI is not used to determine the final classification: after confirmation, the operome applies versioned deterministic rules to the confirmed facts.
5. Your responsibility for facts and intended purpose
Results depend on the product, feature, deployment, role, intended purpose and other facts you declare. You are responsible for checking them against contracts, technical documentation, deployment settings and actual use. A change in model, feature, customer, territory or intended purpose can require a new assessment.
6. GPAI models and presets
The classifier distinguishes your downstream AI system from an underlying reusable model. Using Claude, a GPT model, Gemini or another general-purpose model does not by itself mean that you are the provider of that underlying model, and it does not replace the classification of what your own product does.
Presets are convenience assumptions at a model-family level. They are not statements by Anthropic, OpenAI, Google or another model provider; they are not warranties about a particular model or version. You must confirm the exact model, contractual role, current provider documentation, training information and any formal European Commission designation.
7. Training compute
The FLOPs field concerns cumulative compute used to train the underlying model. It does not concern the number of API calls or tokens, inference usage, hosting expenditure, or compute used merely to run the model inside your application. Leave it unknown unless reliable information for the exact underlying model is available.
8. Interaction records, IP addresses and email
The classifier maintains a detailed operational and security audit trail from the start of a session. It records values entered into fields, answers and options selected, interface actions, proposed and confirmed facts, extraction and classification events, client and server timestamps, the originating IP address, browser user-agent information, page path, source URL and evaluation identifier where available. This applies to both website-led and manual questionnaires.
Before an email is supplied, events are linked to an opaque browser-session identifier. If you choose to unlock and register a result, you authorise SynapseLayer to associate the submitted email with the events already recorded for that browser session, and to store the email, declared facts, computed result and evaluation identifier. A notification containing the submitted email, confirmed facts and result may be sent to the classifier administrator, who may contact you about that compliance question.
Audit records are available only to authorised classifier administrators through the protected administration area and are used for investigation, service security, abuse prevention, support and improvement of the questionnaire. They are retained only for as long as reasonably necessary for those purposes and applicable legal obligations. Do not enter confidential, special-category or other sensitive personal data.
After the first unlock, an opaque, secure browser cookie is set for up to 180 days. It contains no email address and is used only to recognise that browser, retrieve the previously supplied email on the server and save a revision after you explicitly select “Review or change these answers”, without asking for the email again. Starting a separate assessment does not silently reuse that email. Clearing browser cookies or using another browser will require a new email unlock.
A permalink may be accessible to anyone who receives it; do not register or share facts you are not authorised to disclose. Submission and administrator-notification traffic may be throttled to protect the service and mailbox from abuse.
9. No reliance; independent review required
Do not rely on a result as the sole basis for launching, selling, procuring, deploying, suspending or withdrawing an AI system, or for making a filing or representation to an authority or third party. Obtain independent review from appropriately qualified legal and technical professionals before taking action.
10. Disclaimer and liability
To the fullest extent permitted by applicable law, the classifier, extractions, presets, traces, exports and permalinks are provided “as is” and “as available”, without warranty of accuracy, completeness, currency, availability, fitness for purpose or any legal or regulatory outcome.
To the fullest extent permitted by applicable law, SynapseLayer excludes liability arising from or connected with use of, inability to use, or reliance on the classifier or its outputs, including regulatory action, compliance cost, rework, delay, loss of data, revenue, profit, opportunity or goodwill. Nothing excludes or limits liability that cannot lawfully be excluded or limited.