AI Literacy Policy

1. Purpose & Scope

Article 4 of Regulation (EU) 2024/1689 (“EU AI Act”), in force since February 2, 2025, requires that providers and deployers of AI systems take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf. This obligation considers the technical knowledge, experience, education and training of the relevant persons, and the context in which the AI systems are to be used.

This policy establishes how UBava OÜ (“UBava”) fulfills that obligation as both a provider (of HiveGuard autonomous agents and proprietary AI subsystems) and a deployer (of third-party foundation models via the Privacy Relay).

This policy applies to:

• All UBava employees, contractors, and consultants
• Users of the UBava Privacy Relay platform
• Users of HiveGuard autonomous agents
• Third-party integrators consuming UBava APIs

This policy does not replace the separate AI Act Compliance Statement, GDPR compliance documentation, or the VHH Privacy Air-Lock Protocol technical specification. It supplements them.

2. Definitions

These definitions align with Article 3 of Regulation (EU) 2024/1689:

3. AI Systems Operated by UBava

3.1 Privacy Relay (Deployer)

UBava deploys the following third-party foundation models through its VHH Privacy Air-Lock protocol:

No real personally identifiable information leaves the EU boundary. The VHH Privacy Air-Lock (Docker 1) tokenizes PII locally on Hetzner infrastructure in Nuremberg, Germany, before any query reaches a third-party model.

3.2 HiveGuard Autonomous Agents (Provider)

All agents operate under the Bee Dance Protocol (BDP) for inter-agent communication and are bound by jurisdiction-aware autonomy constraints — no agent may take action that violates the law of its operating jurisdiction.

3.3 Proprietary AI Subsystems (Provider)

4. User Literacy Requirements

Before using any UBava AI system, users must understand and acknowledge the following:

4.1 AI Output Limitations

• AI systems can produce incorrect, incomplete, or misleading outputs. This includes factual errors (“hallucinations”), outdated information, and responses that reflect biases present in training data.
• AI outputs are not verified facts. They are probabilistic text generation. Users must apply their own judgment.
• AI outputs should never be the sole basis for legal, medical, financial, or safety-critical decisions. Professional advice from qualified humans is required for such decisions.

4.2 Data Handling Transparency

• The Privacy Relay processes synthetic data only. The VHH Privacy Air-Lock tokenizes all PII before queries leave UBava infrastructure. Third-party AI models never see real personal data.
• Tokenization is not anonymization. The mapping between real and synthetic data exists within UBava’s secure infrastructure and is governed by GDPR data processing agreements.
• Users should not attempt to embed sensitive data in ways designed to bypass PII detection (e.g., encoding PII in base64, splitting identifiers across multiple messages).

4.3 Autonomous Agent Operations

• HiveGuard agents operate autonomously within defined boundaries. They can gather information, analyze patterns, and in the case of the Fixer, take corrective action — all without per-action human approval.
• Human oversight is maintained through the approval queue. Significant actions, new findings, and anomalies are escalated to human operators before execution.
• Agent lifecycle states (DORMANT, RUNNING, HIBERNATING, PAUSED, FROZEN, ESCALATED, RESTING, DECOMMISSIONED) determine what actions an agent can take. Users operating agents must understand these states.

4.4 User Responsibility

• Users retain full responsibility for how they use AI outputs. UBava provides tools; users make decisions.
• Users must not represent AI-generated content as human-authored in contexts where such disclosure is required by law or professional standards.
• Users must report suspected AI system malfunctions, harmful outputs, or security incidents to info@ubava.ee.

4.5 Disclosure Mechanism

Users are informed they are interacting with AI systems through:

• Platform interface labeling identifying AI-generated responses
• API response metadata indicating the model used
• Terms of Service (Section 9.2) disclaiming AI accuracy guarantees
• This policy document, available at /legal/ai-literacy

5. Staff Training Requirements

All UBava staff who develop, deploy, operate, or support AI systems must complete the following:

5.1 Mandatory Training (All Staff)

5.2 Role-Specific Training

5.3 Training Records

Training completion is documented with date, participant, and topic. Records are retained for the duration of employment plus 3 years, in accordance with Estonian labor law and AI Act accountability requirements.

6. Prohibited Uses

The following uses of UBava AI systems are prohibited. These align with Title II (Prohibited AI Practices) and Annex III (High-Risk AI Systems) of Regulation (EU) 2024/1689:

6.1 Absolutely Prohibited (Article 5)

• Social scoring — Using AI outputs to evaluate or classify individuals based on social behavior or personal characteristics, leading to detrimental treatment
• Subliminal manipulation — Using AI to deploy subliminal techniques that materially distort behavior in ways likely to cause harm
• Exploitation of vulnerabilities — Targeting specific groups (age, disability, social/economic situation) with AI-driven manipulation
• Real-time remote biometric identification in publicly accessible spaces for law enforcement purposes (except narrow exceptions not applicable to UBava)
• Emotion recognition in workplace or educational contexts
• Untargeted scraping of facial images from the internet or CCTV to build facial recognition databases

6.2 High-Risk Domain Restrictions (Annex III)

UBava AI systems must not be used as the primary decision-making system in the following domains without separate high-risk compliance assessment and registration:

• Employment and worker management (hiring, termination, task allocation, performance monitoring)
• Credit scoring and financial creditworthiness assessment
• Access to essential private or public services and benefits
• Law enforcement (individual risk assessment, polygraph, evidence evaluation)
• Migration and border control (risk assessment, visa processing)
• Administration of justice (judicial research for applying law to facts)
• Education (admissions, assessment, proctoring)

Users attempting to deploy UBava systems in these domains must contact legal@ubava.ee for a conformity assessment before proceeding.

6.3 UBava-Specific Prohibitions

• Using HiveGuard agents to conduct surveillance on individuals without lawful basis
• Circumventing the PII tokenization pipeline to expose personal data to third-party models
• Using AI outputs to generate spam, phishing content, or disinformation
• Deploying agents with jurisdiction-aware constraints disabled

7. Review Schedule

Version history is maintained in Git. All changes to this document are tracked with commit messages referencing the change rationale.

8. Contact

For questions about this policy, AI literacy training, or to report concerns about AI system use:

• General inquiries: info@ubava.ee
• Legal & compliance: legal@ubava.ee
• Security incidents: See Incident Response documentation

This document fulfills the AI literacy obligation under Article 4 of Regulation (EU) 2024/1689 (EU Artificial Intelligence Act). It is a living document maintained by UBava OÜ and updated as required by Section 7 above.