Effective Date: 11 August 2026 · Version: 1.0
This notice explains how AI is used in Slides by Karya. It is published to meet Article 50 of the EU AI Act (transparency obligations, applicable from 2 August 2026) and because you are entitled to know when you are dealing with a machine.
1. You are interacting with AI
Slides by Karya is an AI system. These features are AI-powered:
| Feature | What the AI does |
|---|---|
| Deck generation | Writes and designs slides from your brief and source material |
| Outline | Proposes a structure before generation |
| Copilot | Edits slides in response to your instructions in natural language |
| Image generation | Creates images and visual elements |
| Style direction | Interprets a reference to set visual direction |
The copilot is an AI assistant, not a person. It is labelled as such in the interface.
2. Decks are AI-generated content
Content produced by the Service is artificially generated. Under Article 50(2) of the EU AI Act we mark synthetic content in a machine-readable format, and under Article 50(4) we require disclosure of AI-generated content published to the public on matters of public interest.
In practice:
- Shared decks carry a visible "AI-generated" label.
- Exported and generated files carry machine-readable provenance metadata identifying the content as AI-generated. *(Art 50(2) marking obligation — see §6 for status.)*
- You must not remove or obscure these markings, and you must not present AI-generated content as human-authored where that would mislead or is prohibited by law.
3. How the AI runs
| Purpose | What runs it | Location |
|---|---|---|
| Text, deck drafting, copilot | An open-weights large language model, run by a third-party inference provider | United States |
| Images | An image generation model, run by a third-party provider | United States |
We run the open-weights model on a US-based host rather than using the model author's own API, so that your content is handled under SOC 2 Type II assurance, contractual zero data retention, and enforceable EU and UK data-protection terms.
Models and providers may change to improve quality, cost, latency, or availability. We identify our current providers by name to customers on request — support@the-karya.com. Business customers under our Data Processing Addendum receive 30 days' notice of any change, with a right to object.
4. What the AI does not do
- It does not make decisions about you. No scoring, profiling, eligibility assessment, or automated decision-making with legal or similarly significant effects. It generates presentation content at your instruction.
- It does not train on your content. See Privacy Policy §4.
- It does not do emotion recognition or biometric categorisation.
- It is not a high-risk AI system under Annex III of the EU AI Act. It is a content-generation tool, not a system used for employment, education, credit, law enforcement, or access to essential services. If we ever build something that falls into a high-risk category, we will say so and meet the additional obligations before shipping it.
5. Limitations you should know about
AI output is probabilistic. It can be:
- Wrong — plausible-sounding facts, figures, dates, and citations that are fabricated. Verify anything that matters.
- Inconsistent — the same prompt can produce different results.
- Derivative — models are trained on large datasets and may produce output resembling existing work. We make no claim that output is original or free of third-party rights.
- Biased — trained on human-generated data, models reproduce patterns in it.
- Out of date — models have a training cutoff.
Review output before you rely on it, publish it, or present it. Do not use it in medical, legal, financial, tax, employment, safety-critical, or regulatory contexts without qualified human review.
6. Where we stand on the EU AI Act
Article 50 of the EU AI Act sets transparency obligations for AI systems. This is what we do:
| Obligation | What we do |
|---|---|
| Tell you when you are interacting with AI | The copilot is labelled as an AI assistant |
| Disclose AI-generated content | Shared decks carry a visible "AI-generated" label |
| Prohibit undisclosed deepfakes | Prohibited outright by our Acceptable Use Policy |
| Provide this information clearly | This notice |
Machine-readable marking. Article 50(2) also requires synthetic content to carry markings a machine can detect, not just a human. That obligation applies from 2 December 2026. Our shared decks currently carry a visible label; we are adding embedded provenance metadata to generated images and exported files before that date.
7. Questions
support@the-karya.com · See also Privacy Policy and Terms §11.