Compliance
Compliance for AI that actually runs processes
Swiss and European rules do not ask whether AI is clever. They ask who is accountable, what was decided, and whether you can prove it.
Compliance for AI that actually runs processes
ÆTERION BRAIAN produces the evidence European supervision expects. Each process carries a risk classification, named human oversight, a data-protection basis and a versioned record of every decision. Documentation required by the EU AI Act, the General Data Protection Regulation (GDPR) and the Swiss Federal Act on Data Protection (nLPD) comes from the runtime. Not from manual work before an audit.
Frameworks we map to
- EU AI Act
- Risk classification per process, documented human oversight, technical documentation, logging and post-market monitoring produced by the runtime.
- GDPR
- Lawful basis recorded per process, data minimisation through redaction, retention policy, records of processing and data-subject request support.
- Swiss FADP (nLPD)
- Swiss data residency, processing records and controller/processor responsibilities reflected in platform configuration.
- ISO/IEC 42001 and 27001
- The platform is built around these control sets; certification of the ÆTERION management system is in progress and the status is published on the trust page.
What does the runtime produce?
- A register of AI use cases with owner, purpose, risk class and status.
- Evidence per run: inputs, decision, model version, policy version, approver.
- Human oversight records: who reviewed, what they changed, why.
- Incident and deviation records with resolution and follow-up.
- Exportable documentation packs for internal audit and supervisors.
Responsibilities
ÆTERION provides the platform, its controls and its documentation. Your organisation remains responsible for the purpose of each process, for the lawful basis of the data it uses and for the decisions it takes. The platform makes that responsibility exercisable instead of theoretical.
Carol and employee data
Last updated: 20 September 2026
Carol's personalisation involves employees' personal data. Under the GDPR and the Swiss FADP, that processing requires its own lawful basis — separate from the basis of the underlying business process — and employees have the right to be informed: which data is used, for what purpose, for how long and who to contact. The controller for that processing is your organisation; the platform records the basis and supports the information and access duties.
What employee data does Carol personalise on?
Carol personalises on work context, not on the person. Specifically: the employee's role and team, the permissions granted to them, the systems they are authorised to access, the processes they are involved in, and their interaction history within Carol. Carol does not process performance evaluations, HR records, private communications, or behavioural indicators. It does not build a behavioural profile of the employee, and it does not infer traits, productivity or engagement. Personalisation is bounded by the same permission model that governs every other part of BRAIAN: Carol cannot surface information the employee is not authorised to see.
What happens to the personalised instance when an employee leaves the company?
The personalised instance is deactivated when the employee's account is deactivated. Carol stops operating immediately. Personalisation data and conversation history are deleted within 30 days of deactivation. Audit records of business actions remain. If Carol was used to start a supplier onboarding or approve an expense, that action stays in the audit log under the organisation's business record retention policy — it is a record of a business decision, held on a separate legal basis from personalisation data. The two are deliberately separate: personal data is deleted, business records are preserved.
Are voice recordings retained? For how long and where are they processed?
Voice input is transcribed to text in real time and the audio is discarded immediately after transcription. Audio recordings are not stored. Transcripts are retained for 30 days and are then deleted. Retention is configurable per deployment and can be set to zero. All processing — transcription and inference — takes place within Switzerland or the EU. Voice data is not transmitted outside the configured region and is not used to train any model, ours or a third party's.
Can Carol be used to evaluate employee performance?
No. Carol produces no metrics on individual employees: not usage frequency, not question volume, not response quality, not any derived indicator. Carol is not an evaluation system and is not designed to become one. No report, export or API exposes per-employee activity data. Process metrics in BRAIAN measure processes, not the people who run them.
Can a manager access a team member's Carol conversations?
No. Carol conversations are visible only to the employee who had them. Managers, administrators and ÆTERION personnel have no access to conversation content. What is recorded in the audit log is the action, not the conversation: if a conversation results in a business process being started, the audit log shows that the process was initiated, by whom and under which policy — not what was said. Access to conversation content is possible only through a documented exceptional-access procedure, which requires written authorisation from the organisation's data controller and is itself logged. This is a design constraint, not a configuration option. Swiss labour law prohibits systems designed to monitor employee behaviour, and Carol is built to that constraint.
FAQ
Is the platform certified?
ÆTERION operates to ISO/IEC 27001 and 42001 control sets; formal certification is in progress. We publish the current status rather than implying a certificate we do not yet hold.
Does BRAIAN classify our processes automatically?
It guides the classification and records it, including the reasoning. The decision stays with your compliance owner.
Can auditors read the evidence directly?
Yes. Read-only auditor access can be granted to the register and to traces, scoped by the control plane.
Prepare your AI documentation once
We review one process with your compliance team and show the evidence the platform produces.
