EU AI Act & Responsible AI

SARAMA is designed to keep people informed and in control when AI employees carry out work. This page explains the product controls we provide and the responsibilities that remain with each customer.

The regulatory context

Regulation (EU) 2024/1689—the EU AI Act—uses a risk-based framework for providers and deployers of AI systems. It entered into force on 1 August 2024 and applies in stages:

• 2 February 2025 — already in force: the prohibited practices in Article 5, and the Article 4 AI-literacy duty on providers and deployers. If you use SARAMA today, the AI-literacy obligation applies to you now, not in 2026.
• 2 August 2025: obligations for providers of general-purpose AI models began to apply, subject to the transitional rules for models already on the market.
• 2 August 2026: the general application date, including the Article 50(1), 50(3) and 50(4) transparency duties. Article 50(2) applies to newly placed generative systems from this date; systems placed on the market before this date have a transitional period until 2 December 2026.
• 2 December 2027: the delayed Chapter III provider obligations for high-risk systems under Article 6(2) and Annex III.
• 2 August 2028: the delayed Chapter III provider obligations for high-risk systems under Article 6(1) and Annex I.

Article 50 has four limbs that matter to a platform like this one: people must be told when they are interacting with an AI system unless it is obvious (50(1)); synthetic audio, image, video and text must be machine-readably marked as artificially generated (50(2)); deep fakes must be disclosed (50(3)); and AI-generated or AI-manipulated text published to inform the public on matters of public interest must be disclosed unless it underwent human review with someone holding editorial responsibility (50(4)).

Official sources: consolidated EU AI Act on EUR-Lex, Regulation (EU) 2026/1744, and European Commission transparency guidelines.

How SARAMA supports responsible use

• Visible AI identity: AI agents and agentic teams are presented as AI inside the platform.
• Human control: customers can run work automatically, co-create with an agent, or require approval before consequential steps.
• Permission boundaries: agents operate within organisation and user permissions.
• Auditability: platform activity and consequential actions can be recorded for review.
• Provider choice: customers choose and connect their own supported AI model providers.
• Data separation: customer data remains isolated by organisation-level access controls.

Our roles, and where they change

SARAMA does not train its own models. Every generation runs on an AI model the customer connects with their own provider account and API key, so the model providers are the providers of those general-purpose models. For the AI features built on top—agents, agentic teams, the chat assistant and content generation—we act as the provider of that AI system, and the customer acts as its deployer. Deployer duties, including Article 50(1) and 50(4) disclosure to the people a customer reaches, sit with the customer, because only the customer knows the context, audience and purpose of a given workflow.

Transparency built into the product

These are mechanisms SARAMA implements, not intentions:

• AI chat is disclosed before you engage (Art. 50(1)): where a customer runs an AI chat assistant on their site, the AI-interaction notice appears in the first layer of the consent banner—readable before the accept button, not hidden behind a “manage” screen.
• Content drafted or translated by our AI carries the Art. 50(4) disclosure: where an item is written, translated or SEO-optimised through the SARAMA AI assist, that provenance is recorded on the item. The marker only ever accumulates—it is never cleared by a later human edit, because “some of this was AI” stays true. When such an item is published through the CMS, the disclosure is appended by the renderer itself, so every delivery path carries it rather than only the page we remembered to update. Provenance is written in that one place: an item composed or imported outside the AI assist path carries no marker, and the renderer has nothing to disclose.
• Editorial review is respected: Article 50(4) does not require disclosure where a human took editorial responsibility. Customers can override the default per site, and SEO-only assistance is not labelled by default, since metadata is not content published to inform the public.
• Generated media carries descriptive alt text, not the raw prompt used to make it.
• Approval gates: a workflow can be set to pause for human approval before a consequential step, and the run history records who approved what.

One current limit: SARAMA does not add its own machine-readable provenance marker to generated media that arrives without one. Where a model provider embeds provenance metadata such as C2PA and the delivery pipeline does not transform the file, SARAMA preserves it. Customers must not rely on SARAMA alone to satisfy Article 50(2) for unmarked synthetic media. For generative systems placed on the market before 2 August 2026, Regulation (EU) 2026/1744 provides a transition until 2 December 2026; this does not postpone the other Article 50 duties.

Customer responsibilities

The legal role and obligations of a customer depend on the use case, sector, territory and people affected. Customers must:

• assess whether a planned use is prohibited, high-risk or subject to specific transparency duties;
• tell people when they are interacting with an AI system where required;
• review AI-generated output before using it for decisions that may materially affect a person;
• provide appropriate human oversight, staff instructions and AI literacy;
• respect privacy, intellectual-property, consumer-protection, employment and sector-specific rules;
• avoid using SARAMA for unlawful discrimination, manipulation, social scoring or other prohibited practices.

Not a blanket compliance certificate

No general-purpose business platform can determine the legal classification of every workflow a customer creates. This page is product information, not legal advice or a statement that every customer configuration is compliant. Obtain qualified advice for regulated or high-impact uses.

Questions and documentation

For AI Act, transparency or responsible-AI questions, contact impact@angad.swiss. We can provide product and security information relevant to your assessment.

Last reviewed: 1 September 2026