Уведомление и условия использования AI

1. What the AI feature does

VIEWMIUM AI provides educational explanations and navigation based on a user’s question and selected published atlas content. The retrieval-based feature may send the question, locale, structured atlas context and retrieved published text to a contracted model provider. It does not send raw DICOM, NIfTI, video or image binaries.

You are interacting with an AI system. Responses are generated probabilistically and may be inaccurate, incomplete, outdated or fabricated. Use AI only for education, do not enter patient information, and verify important statements against cited sources and qualified professional guidance.

2. First-use disclosure and permission

Before the first request to an external AI provider, and again after a material provider or data-use change, the interface must identify the feature as AI, name the provider, describe the categories sent and their purpose, disclose retention and training use, link to relevant privacy information and state the no-PHI rule. Where required, the user must affirmatively choose to continue. Declining must not disable unrelated non-AI features.

3. Permitted use

• Ask educational questions about published anatomy, modalities, findings and concepts.
• Request a plain-language explanation, study aid or navigation to cited VIEWMIUM material.
• Use output as a starting point for independent learning and source verification.
• Report an inaccurate, unsafe, biased, private or rights-infringing response.

4. Prohibited inputs and uses

• Do not provide PHI, personal health records, names, dates of birth, record numbers, images or facts about an identifiable patient.
• Do not ask the AI to diagnose, interpret a live study, triage, recommend treatment, determine urgency or replace a clinician.
• Do not use output for an exam, credential or assessment in violation of applicable rules.
• Do not attempt to extract system prompts, secrets, private content, model weights or restricted datasets.
• Do not use automated means to evade limits, generate harmful content, attack the Service or infringe rights.
• Do not represent AI output as independently authored, peer-reviewed or clinically validated.

5. Accuracy, sources and oversight

AI output is not guaranteed. Retrieval can omit relevant context, citations can be mismatched and translations can lose nuance. Where feasible, the interface identifies the source title, contributor, version and link. A source link does not mean the generated statement is endorsed by the source or contributor. Human editorial review applies to published atlas content, not every private AI response.

Safety controls may block, redirect or log high-risk prompts. Emergency-like prompts receive a non-clinical emergency notice. Flagged outputs may enter an internal review process, but are not monitored continuously and no response time suitable for patient care is promised.

6. Provider, retention and training

Before external processing begins, the in-product disclosure identifies the relevant provider or service, the data categories sent, their purpose, and the available information about retention, training use, deletion and limited human access.

Neither contributor medical cases nor private user content may be used to train a VIEWMIUM or third-party model under these Terms. Any future training use requires a separate, specific opt-in and rights review.

7. Intellectual property and feedback

Rights in prompts and output are subject to applicable law, the Terms, source-content rights and provider terms. VIEWMIUM does not warrant that output is unique or non-infringing. Feedback about AI may be used to improve safety and service quality, but not to train a model on private content unless a separate training permission expressly says so.

8. Availability and changes

AI may be rate-limited, unavailable, changed or withdrawn. Provider or model changes must pass privacy, security, sanctions, quality and medical-claims review. Material changes require updated disclosures and, where applicable, renewed permission before data is sent.