AI transparency notice
Last updated 29 August 2026
You are interacting with an artificial intelligence system, not a human being. Erissa writes in first person and forms a long-term memory of you, but there is no person on the other side.
This notice covers our Article 50 transparency duties under Regulation (EU) 2024/1689 (the AI Act) and describes how we govern the system.
Risk classification
Erissa is a general-purpose-AI-based conversational assistant used for personal companionship and organisation. We classify it as a limited-risk system subject to transparency obligations. It is not used for employment, education, credit, insurance, law enforcement, migration or biometric decisions, and it must not be used for any high-risk purpose listed in Annex III.
Prohibited practices we do not engage in
No subliminal or manipulative techniques designed to distort your behaviour, no exploitation of vulnerability, no social scoring, no emotion inference in workplace or education settings, and no biometric categorisation. Erissa may note how you say you feel because you told it so — it does not claim to detect emotions from your face or voice.
How it works
| Element | What happens |
|---|---|
| Models | Third-party large language models, selected per persona. Named in Settings and on the subprocessors page. |
| Memory | Facts are extracted from what you say, scored for confidence and importance, and stored in your account only. |
| Learning | Erissa adapts to you through your stored memory and personality settings. Your conversations are never used to train vendor foundation models. |
| Inference | Erissa may infer things you didn't state explicitly. This is off-switchable under Controls and every inference is labelled 'I worked it out'. |
Known limitations
Language models can be confidently wrong, invent details, misremember context, reflect biases present in their training data, and give inconsistent answers to the same question. Erissa is not a source of medical, legal, financial or crisis advice. Check anything that matters.
Human oversight and your control
You are the oversight. You can read, edit, pin, lock, reject or delete every memory; view the full audit log of what changed and whether you or the system did it; pause learning entirely; approve or reject proposed memories; adjust six behaviour dials; and export or erase everything. We review reported outputs and can disable behaviours globally.
Synthetic content
Text Erissa produces is AI-generated. If you republish it elsewhere, label it as such — that is your obligation as a deployer under the AI Act.
Reporting a problem
Report harmful, biased or unsafe output to security@erissa.ai. We log incidents, respond within five working days, and escalate to the relevant market surveillance authority where required.
Caan Consultancy Ltd, a company registered in England and Wales, company number 09178365. Registered office: Lytchett House, 13 Freeland Park, Wareham Road, Lytchett Matravers, Poole, Dorset BH16 6FA, United Kingdom. Erissa is a trading name of Caan Consultancy Ltd.
Contact us: privacy@erissa.ai for data requests and your rights, support@erissa.ai for general questions and complaints, and security@erissa.ai for responsible disclosure. We answer data requests within one month, as required by the UK GDPR and EU GDPR.