Trust center
Caire and the EU AI Act
Which parts of Caire use AI, how we assess them under the AI Act (EU) 2024/1689 and who has the final say.
No part of Caire uses any of the AI practices the Act prohibits.
A planner approves every schedule before it is first published.
Caire calculates no ratings, scores or performance measures for individual employees.
Our assessment in brief
- EirTech AB (Caire) is the provider of the AI features in Caire. The organisation that uses Caire is the deployer under the Act.
- Fråga CAIRE and the AI-prepared drafts are limited-risk AI systems. The transparency requirements in Article 50 apply to them.
- The optimisation that builds schedules, routes and replacements follows rules and weights set by people. It does not learn from personal data. Under the European Commission's guidelines on what counts as an AI system, we assess that it is not an AI system within the meaning of the Act.
- The optimisation allocates work between employees. The Act counts this as high risk when an AI system does it (Annex III point 4). We therefore claim no exemption for it and work with human oversight, logging and documentation.
- Caire does not assess who is entitled to care. Municipal care decisions are input to the planning.
The parts of Caire and how they are assessed
Every part that uses AI or makes automatic decisions, what it does and who has the final say.
| Part | What it does | Who decides | Our assessment |
|---|---|---|---|
| Schedule and route optimisation | Proposes schedules, routes and shifts | The planner publishes | Rule-based optimisation, not an AI system. Run with human oversight and logging |
| Replacement after a sick call | Proposes who takes a shift | The planner, or automatically when a colleague has said yes | Rule-based optimisation. Allocates work, so the same controls as above |
| Recommended visit times | Shows suitable times for a new visit | The user chooses | Rule-based optimisation, not an AI system |
| Fråga CAIRE in Caire | Answers questions about the planning and prepares proposals | The user confirms | Limited-risk AI system |
| Fråga CAIRE on caire.se | Answers general questions about Caire | Nothing is changed | Limited-risk AI system |
| Draft care plans and notes | Drafts from speech and records | The employee reviews and saves | Limited-risk AI system |
| Connections for AI assistants (MCP) | Lets the organisation's own AI assistant read and prepare proposals | The user confirms | Limited-risk AI system |
Human oversight
- Fråga CAIRE changes nothing on its own. The assistant prepares proposals, and a proposal is carried out only after the same user has confirmed it.
- Draft care plans and structured visit notes are saved only after an employee has reviewed them.
- Recommended visit times only fill in the form. The user chooses and saves.
- Proposals, approvals and publications are kept in a hash-chained decision log. Changes from Fråga CAIRE and AI assistants connected to Caire are logged with where they came from.
What can happen automatically
After the first publication there are two automatic paths. Both follow rules the organisation controls.
- When an employee falls ill and a colleague has said they can take the shift, the replacement can be approved automatically. The organisation can switch this off.
- In organisations with automatic planning, the times in a published schedule can be adjusted when something happens during the day. The same employee keeps the visit, and the visit stays within its allowed time window.
- Every automatic change is checked first. If a visit would be left without staff, go to someone who is not approved or fall outside its time window, the change becomes a proposal to the planner instead.
- Automatic publications are marked in the log as automatic, together with the rule that applied.
Employees and clients
- The schedule is built from skills, availability, continuity, preferences, workload and travel time.
- Caire uses no emotion recognition and identifies no one with biometrics. Unlocking with Face ID or Touch ID happens on the phone, and Caire receives no biometric data.
- Location data is not used to assess employees. The client sees the employee's position only during a visit and only when both have consented.
Transparency
- Fråga CAIRE states clearly that it is an AI assistant, both in Caire and on caire.se, and its answers are marked as AI-generated.
- Text prepared by AI is shown as a proposal and marked in the interface.
- Visit notes structured by AI and answers Fråga CAIRE produces with AI are also marked in a machine-readable way, so other systems can tell the text is AI-generated. Prepared answers written by us are not marked.
- The chat on caire.se gives general product information. It does not accept personal data and changes nothing in Caire.
Language models and AI literacy
- AI features for signed-in users run in Azure Sweden Central. Caire trains no general-purpose AI models of its own.
- Your data is not used to train generative models.
- Everyone who works at Caire gets a security and data protection briefing, covering AI tools, when they start and at least once a year.
When the rules apply
The dates after the EU amending regulation 2026/1744. We update the assessment when the rules or the Commission's guidelines change.
| Date | What starts to apply |
|---|---|
| 2 February 2025 | Prohibited AI practices and AI literacy |
| 2 August 2025 | Requirements for providers of general-purpose AI models |
| 2 August 2026 | Transparency requirements for AI systems that interact with people |
| 2 December 2027 | Requirements for high-risk systems in Annex III, including employment |
Questions about AI in Caire: support@caire.se
This page describes Caire's own assessment of its own service. It is not legal advice, and each organisation makes its own assessment of how it uses Caire.