EU AI Act Compliance Statement

1. EU GPAI Deployer Status

Intended is a deployer, not a provider, of general-purpose AI (GPAI) models. Authority decisions and policy evaluation never use a GPAI model. For a customer's requests, an AI model is used only in two optional features, both off by default and turned on only by the customer: AI-assisted classification of plain-English requests (OpenAI or Anthropic, using the customer's own API key and under the customer's agreement with that provider) and AI policy drafting (a model Intended hosts in its own infrastructure, producing suggestions that a person reviews before anything is saved). Separately, Intended may train and run its own models, hosted in its own infrastructure, on de-identified or redacted data for product improvement, as described in Privacy Policy Section 4B; no customer data is sent to a third-party AI provider for that work. For these uses Intended maintains documentation, implements safeguards, and respects data protection. EU Legal Representative: designation in progress; until the representative is published, contact eu-privacy@intended.so for all EU AI Act representative matters. For EU AI Act questions, contact eu-privacy@intended.so.

2. Overview

Intended is an authorization and governance layer for AI agents. We are committed to compliance with the EU Artificial Intelligence Act (Regulation (EU) 2024/1689).

3. System Classification

Intended's core authorization engine is a deterministic, rule-based policy evaluation system. It does not use machine learning, neural networks, or statistical inference to make authorization decisions. Under Article 3(1) of the AI Act, which defines "AI system" as a machine-based system that infers from inputs to generate outputs such as predictions, recommendations, or decisions, Intended's policy engine does not meet this definition because:

• Policy evaluation is fully deterministic — identical inputs always produce identical outputs
• Risk scoring uses explicit, human-configured rules and thresholds, not learned models
• Authorization decisions (ALLOW, DENY, ESCALATE) are the result of rule evaluation, not inference
• All decision logic is transparent, auditable, and configurable by the customer

4. Intent Classification Component

By default, Intended's intent classification component is deterministic: it sorts requests into structured intents using versioned rules, and every decision receipt shows which classifier and rule-set version made the decision. If a customer turns on AI-assisted classification, an AI model suggests the classification for plain-English requests only, as a preprocessing step. The authorization decision itself is always made by the deterministic policy engine, not by the classification component. Intended may use machine learning, including models it trains and hosts itself, on de-identified or redacted data to improve classification (see Privacy Policy Section 4B), but those improvements reach the default decision path only as new, versioned deterministic rule sets.

5. Human Oversight

Intended is designed with human oversight as a core principle:

• High-risk actions are automatically escalated for human review
• All automated decisions can be overridden by authorized human operators
• Complete audit trails provide full transparency into every decision
• Fail-closed architecture ensures that system failures result in denial, not unauthorized approval

6. Customer Guidance on High-Risk Use Cases

Customers deploying Intended to govern AI agent actions in the following domains should be aware that their overall AI system (not Intended alone) may be subject to Annex III high-risk classification:

• Employment and worker management (Section 4)
• Access to essential private services, including financial services (Section 5)
• Law enforcement (Section 6)
• Migration, asylum, and border control (Section 7)
• Administration of justice (Section 8)

7. Our Commitments

• We maintain technical documentation describing our system's architecture and decision logic
• We provide transparency tools (audit trails, evidence bundles) to support customer compliance
• We cooperate with customers in conducting AI impact assessments
• We monitor evolving guidance from the European AI Office

8. Contact

For questions about Intended's AI Act compliance, contact compliance@intended.so.