AI SYSTEMS NOTICE

AI Notice & Algorithmic Transparency

The ARTEMIS Platform — operated by CAI Technology S.R.L.

Aligned with Regulation (EU) 2024/1689 (EU AI Act) — Articles 13, 22, 50, 52 — and GDPR Articles 13, 22.

Version: v1.0 ARTEMIS • Published on: 2026-05-06 • Effective from: 2026-05-06

1. Overview

• ARTEMIS chatbot: AI system interacting with natural persons — disclosure obligation under Article 50(1) AI Act.

• Scoring engine: AI system with a limited purpose — minimal/limited risk, no automated legal effects (Article 22 GDPR).

• Report generator: artificially generated content — labeling obligation (Article 50(2) AI Act, phased application).

• LLM50 audit: testing tool — by its nature, not a "high-risk" AI system; produces evaluation content intended exclusively for the Client.

2. ARTEMIS — Conversational assistant

Under Article 50(1) AI Act, CAI Technology S.R.L. expressly informs the User that they are interacting with an AI system. The technical banner is permanently displayed in the ARTEMIS chat interface.

3. Scoring and risk engine

Scoring algorithms calculate the compliance score (percentage), the risk level (CRITICAL / HIGH / MEDIUM / LOW), and prioritize recommendations based on Questionnaire answers. The logic combines:

• explicit rules (response ↔ score mapping, weighting per regulatory framework);

• optional statistical models for clustering and sectoral benchmarking.

Under Article 22 GDPR, scores do NOT produce decisions with legal effects on the User. Decisions remain the User's responsibility. The User may at any time request:

• an explanation of the scoring methodology (at dpo@caitech.ro);

• human intervention (review of the Report by a CAI Technology S.R.L. specialist, on Professional/Enterprise Plan);

• challenge an individual recommendation and add their own remarks.

4. Report generator and recommendations

PDF Reports are generated by combining: (a) data declared by the User; (b) CAI Technology S.R.L.-pre-approved scales and texts; (c) LLM-generated content for personalizing recommendations. Under Article 50(2) AI Act, artificially generated content must be labeled — Reports include a footer with the note "Partially generated with AI assistance" and the exact date of generation.

5. LLM50 audit module

LLM50 is a testing tool for clients who develop or operate their own LLM models. How it works:

• The Client provides an inference endpoint for the model under test (or credentials).

• The Platform sends 50 predefined tests (5 categories: Safety, Security, Privacy, Bias, Reliability).

• Responses are evaluated automatically and manually (mix); an audit report is generated with a score per category, identification of vulnerabilities, and recommendations.

The audit is informational and does NOT constitute formal certification. For purposes of CE marking under the AI Act (high-risk systems), the Client must approach a notified body under Article 43 AI Act.

6. Users' rights regarding AI systems

In connection with the Platform's AI systems, the User has the following specific rights (in addition to the general GDPR rights):

• the right to clear and transparent information about AI use (Article 13 GDPR + Article 50 AI Act);

• the right to request human intervention before a scoring-based decision is taken (Article 22(3) GDPR);

• the right to challenge an automated decision (Article 22(3) GDPR);

• the right to express their point of view on the outcome (Article 22(3) GDPR);

• the right to refuse processing for predictive marketing (Article 21(2) GDPR);

• the right to meaningful information about the logic involved (Article 13(2)(f) GDPR);

• the right not to be misled about the AI nature of an interlocutor (Article 50(1) AI Act).

All these rights are exercised either through the Platform's features or by e-mail at dpo@caitech.ro.

7. Use of User data for AI model training

By default, CAI Technology S.R.L. does NOT use User data for training third-party LLM models. Where contractually agreed (Enterprise customers, for fine-tuning their own model on their own data), the technical terms, retention periods, and segregation guarantees are detailed in the specific DPA.

8. Bias, fairness, and quality

CAI Technology S.R.L. exercises due diligence to minimize bias risks in the AI systems used, by:

• choosing foundation models evaluated by public fairness benchmarks;

• periodic auditing of outputs for systemic deviations (internal red teaming);

• monitoring user complaints and withdrawing/recalibrating problematic modules;

• external audits on request for Enterprise clients.

9. Security of AI systems

AI systems are protected through additional measures:

• prompt filtering against prompt-injection / jailbreak attacks;

• rate limiting per Account and per IP;

• anomaly monitoring on the volume and pattern of queries;

• logging of sensitive queries for audit;

• internal data-use policy with strict access on a "need-to-know" basis.

10. AI Act technical documentation

CAI Technology S.R.L. maintains, under Article 11 and Annex IV AI Act, internal technical documentation for the relevant AI systems. Relevant portions can be made available to competent authorities (AI Office, national supervisory authority) on motivated request. For Enterprise clients, a summary of the documentation may be provided under NDA.

11. Changes

This Notice is updated upon: (a) integration of a new LLM provider; (b) significant changes to scoring algorithms; (c) phased entry-into-force of AI Act obligations (calendar 2025–2027); (d) relevant national legislative changes. Changes are communicated by e-mail and via in-Platform banner.

12. Contact

CAI Technology S.R.L.

Registered office: Str. Victor Brauner, bloc 1, sc. 4, ap. 5, sector 3, București, cod poștal 032621, România • VAT ID: RO50512457 • Companies Register no.: J2024020380005

General e-mail: office@caitech.ro

Support: tehnic@caitech.ro

Data protection contact: dpo@caitech.ro

Security: office@caitech.ro

Related documents. This AI Notice must be read together with the Privacy Policy (GDPR roles + DPA), the Terms and Conditions (AI clauses), the Disclaimer (technological limits), and the Acceptable Use Policy (AI-related prohibitions).