AI Transparency Notice

Version 1.0 · Last updated 13 September 2026 · Canonical URL: https://airhive.ai/legal/ai-transparency

1. Why this notice exists

Airhive GmbH, Eckernwoort 16, 22607 Hamburg, Germany ("Airhive") makes AI Features available under its own name as part of the Airhive Service. Airhive therefore has its own obligations as a provider of an AI system under Regulation (EU) 2024/1689 (the "AI Act"), in particular the transparency obligations of Article 50, which apply from 2 August 2026, and the AI-literacy obligation of Article 4. This notice explains, in plain language, what the AI Features do, which models and providers are involved, what they cannot do, and how Airhive meets these obligations. It forms part of the Terms of Service.

2. What the AI Features do

Airhive uses machine-learning and generative-AI models for the following purposes:

• Text extraction (OCR)converting scanned documents and images into searchable text.
• Classification and namingproposing a document type, a title and a filing location (Hive) for a file.
• Entity and metadata extractionidentifying names, organisations, dates, amounts, reference numbers and similar data in documents.
• Semantic searchfinding documents by meaning rather than exact words.
• Document chatanswering questions about the content of selected documents.
• Summaries and suggestionsgenerating summaries, duplicate detection, clean-up and lifecycle suggestions.
• Agent-assisted organisationapplying instruction sets to propose or, where the Customer has deliberately enabled automatic application above a confidence threshold, to apply organisational changes.

3. Models and providers

All AI processing takes place within the EU/EEA. Customer Content is transmitted to the AI provider per request, only for the enabled feature and only to the extent of the scope configured by the Customer, and is not retained by the provider beyond processing. Neither Airhive nor its providers use Customer Content to train, fine-tune, evaluate or improve any model; see Section 8 of the Terms of Service. Airhive may change or add models or providers within the EU/EEA; changes are announced under Section 11 of the DPA and reflected here.

4. How Airhive meets Article 50 AI Act

Interaction with an AI system (Article 50(1)). Wherever you interact directly with an AI system — in particular in document chat and when reviewing AI proposals — the interface labels the counterpart as AI ("Airhive AI") and the first message of every chat session states that answers are generated by an AI system and may be inaccurate.

Marking of AI-generated content (Article 50(2)). Generated Output — summaries, suggested titles, extracted values, chat answers and organisational proposals — is visibly labelled "AI-generated" in the interface until a user has reviewed and accepted it. Where Generated Output is exported as a file or delivered through the API, it carries a machine-readable marking: exported documents include the metadata field "generator: Airhive AI" and, where the format supports it, a content-credentials manifest; API responses include the field "ai_generated": true and the model identifier used. Accepted and edited output retains a record of its AI origin in the audit log.

Human oversight. AI proposals are pending until a user accepts, edits or rejects them. Automatic application applies only where an Administrator has deliberately enabled it above a confidence threshold, and every automatically applied change can be reversed from the audit log.

5. Limitations you should know about

• Generated Output can be inaccurate, incomplete, outdated, biased or inconsistent, and may present wrong information confidently. Always review it before relying on it.
• Document chat answers are based only on the documents in the selected scope; the model does not have knowledge of documents outside that scope or of events after its training data.
• Extraction of numbers, dates and names can fail on poor scans, handwriting, unusual layouts or languages other than the main European languages.
• Classification and naming reflect patterns in your documents and instruction sets; unusual documents may be misfiled.
• AI Features are not designed for, and must not be used as the sole basis of, decisions with legal or similarly significant effects on individuals.

6. AI literacy (Article 4 AI Act)

Airhive supports the AI literacy of Customers and their users through: this notice; in-product explanations of what each AI Feature does and where its output comes from (confidence indicator, files read, instruction version); the Documentation at https://airhive.ai/docs/ai, which describes appropriate use, review practices and known failure modes; and release notes announcing changes to models or features. Customers deploying Airhive in their organisation remain responsible for the AI literacy of their own staff as deployers.

7. Your role as deployer

If you use Airhive in a professional context, you are a deployer under the AI Act for your own use of the AI Features. Airhive's standard AI Features are not high-risk AI systems within the meaning of Annex III of the AI Act. If you intend to use Airhive in a high-risk context (for example decisions on employment, credit, access to essential services or law enforcement), you must carry out your own assessment and must not rely on Airhive's standard features as the decision-making instrument; see Section 9 of the Terms of Service and Section 5 of the Acceptable Use Policy.

8. Contact

Questions about AI Features and this notice: privacy@airhive.ai or support@airhive.ai.