BLUNTRANSPARENCY

Here we explain where BLUN uses artificial intelligence, what our models King, Voice and Imagine are intended for, where their limits lie and which obligations apply when using them.

1. Recognizing AI and understanding results

BLUN offers AI-supported functions for chat and project work, speech, images, software development and automated workflows. In these functions you interact with artificial intelligence or use results generated by it. Human review of every answer does not take place automatically.

AI can deliver plausible but incorrect answers, omit information, misrepresent sources and adopt distortions from contained patterns. Check results according to their importance. General information and assistance functions do not replace individual professional advice and are not suitable for particularly consequential decisions without further requirements. The statutory rights and the responsibility of BLUN remain in place.

2. Product areas and intended use

King, Voice and Imagine are BLUN's public model names. Chat and Work, Code, Website Builder, Apps, Software, API, Agents and MCP are product and system areas that can use such models. We do not name technical base models here. Functions are only part of the service to the extent actually available and booked.

3. Public model information

These model families are to be distinguished from their individual versions. For each model offered, we document the public version identifier, the as-of date and relevant functional changes:

• King: version identifier and changes: see the change log on docs.blun.ai, as of 13.10.2026
• Voice: version identifier and changes: see the change log on docs.blun.ai, as of 13.10.2026
• Imagine: version identifier and changes: see the change log on docs.blun.ai, as of 13.10.2026

We base published performance claims on tests that have actually been carried out. Technical internals are neither invented nor equated with a mere brand name.

4. Technical basis and transparency

Product names and the underlying technical model basis are different levels. What is decisive for use is the agreed scope of functions, the known limits and the actual data flows. We do not disclose internal architecture details solely through a general product description.

This design does not replace mandatory information. We provide legally required details, rights arising from model and software licenses and information required for customers or authorities. We name the actual recipients of personal data in the section Privacy; they are not hidden behind a vague label such as “AI technology”. A change to the technical foundation must not silently degrade agreed privacy or performance features.

5. Three types of AI information

Information about the interaction: In direct communication with AI, it is clearly explained, at the latest upon the first interaction, that an AI is involved, unless a statutory exception applies. This also applies to communication with end users integrated via API or MCP. The notice must be understandable and accessible; a hidden place in the terms and conditions is not sufficient.

Technical detectability of outputs: For covered generated texts, images, audio and, where applicable, video content, the applicable requirements for machine-readable labeling and detectability must be implemented. The solution used must take into account the content type, technical feasibility and applicable exceptions. A general visible AI notice does not replace the required technical labeling. Source code and other special cases are assessed according to the applicable rules; not every file requires the same watermark.

Disclosure upon publication: Business users must in particular observe the applicable disclosure obligations for covered deepfakes and for certain AI-generated text publications on matters of public interest. A technical metadata item alone does not replace the required understandable disclosure to humans. Statutory exceptions, for example in the case of actual human review and editorial responsibility for certain texts, are assessed on a case-by-case basis. A mere spell check is not a substantive review. Existing provenance labels must not be removed for the purpose of deception.

6. Voice, images and real people

Synthetic voices and images must not be used to deceive about a real person, their consent or an action that actually took place. Required consents and rights must be obtained before processing. The microphone permission of your device does not permit any arbitrary further storage, no biometric analysis and no voice imitation.

In direct speech interactions and realistic-looking synthetic depictions, particular transparency may be required. BLUN does not grant any general permission for biometric identification, categorization or emotion recognition. Such purposes require a separate legal and technical assessment and are excluded outside expressly approved use cases. The stricter child protection and content rules of the usage policy (TERMS, Part C) remain in force.

7. Deployment limits, high risk and prohibited practices

The fact that a feature is generally available does not mean it is approved for every use case. Prohibited AI practices must not be implemented with BLUN. In accordance with applicable law, these include in particular impermissible manipulative practices, the exploitation of particular vulnerability, certain forms of social scoring and prohibited biometric applications. The statutory prohibitions and exceptions are not replaced by this short description.

Regulated or potentially high-risk uses, for example in certain personnel, education, credit, health, authority or security decisions, require an independent assessment of the intended purpose and the statutory requirements. BLUN products are intended for this only if this has been expressly agreed with the necessary measures.

A general chat or a coding tool does not become a high-risk system merely by being called an agent. Conversely, a note “assistance only” does not prevent classification if the actual intended purpose and use meet the statutory criteria. Changes by customers or integrators can affect the allocation of roles.

8. Agents, permissions and human control

Agents operate within the assignment given and the access granted. Text read from a website, file or tool response may neither extend rights nor independently authorize a sending, a purchase or a publication. Authorizations must make the relevant target recognizable and, where necessary, the recipients, content, costs and system.

For high-risk actions, appropriate confirmation, monitoring and a way to abort must be provided. A previously granted automation permission only applies within the expressly described, limited scope. Technical recording of actions must be distinguished from unlimited storage of all content; we describe the retention periods and access rules for this in the section Privacy.

9. Data, memory and learning

Processing an input to generate a response, a voluntary memory function and model training are different purposes. BLUN does not use content for training or fine-tuning without separate explicit consent. A feedback click, connecting a tool or enabling memory is not a blanket consent for this.

The section Privacy explains the actual recipients, retention periods, deletion and your rights. A statement about European platform infrastructure does not automatically mean that services you connect yourself also process exclusively in the EU.

10. Version information, incidents and complaints

Each published product area receives a date of status and a traceable version of its essential information. We handle changes to essential functions, limits or data flows according to the rules of the TERMS and of the Privacy section. A technical model update is no automatic permission to change agreed contractual features.

You can report problematic output, suspected impermissible use cases or erroneous blocks via the reporting channel shown in the product, via the Contact form or report it by email to legal@blun.ai. Only provide the necessary details and no copies of unlawful depictions of abuse. A competent person reviews justified complaints. Statutory complaint and legal remedy channels remain available. Details are set out in the section Reporting channel.

11. Statements and claims about conformity

This description explains products and obligations; it is not a certificate and no blanket guarantee of conformity. We only publish certifications, adherence to codes of conduct, labeling solutions or completed audits if they are actually substantiated. The fact that we do not name specific base models here does not release us from internal documentation and from required information to supervisory authorities and entitled customers.