AI AGENT INTELLIGENCE · EU AI ACT
EU AI Act compliance is no longer a deadline. Article 50 is already in force.
Since 2 August 2026, any AI system interacting with a customer in the EU must make clear that it is an AI. Most organisations running conversational agents can state the policy. Very few can produce the evidence.
WHAT THIS ACTUALLY MEANS
EU AI Act compliance for a customer-facing AI agent rests on three practical duties.
Transparency (Article 50)A person interacting with an AI system must be informed of that fact — in force since 2 August 2026.
TraceabilityThe operator must be able to reconstruct what the system did and when.
Human oversightThere must be a real path for a person to take over. High-risk classification under Annex III, which applies to use cases such as creditworthiness assessment, brings heavier obligations from 2027.
Audited against those dutiesLexic Compass audits AI agents already in production against those duties and returns a Trust Score from 0 to 100 with a signed verdict.
Trust Score 0-100ClearedCleared with ConditionsNot Cleared
It is an independent audit, not a certification: Lexic is not a notified body under the EU AI Act.
What compliance requires of a conversational agent
| Obligation | What must be demonstrable |
|---|---|
| Article 50 — AI disclosure | The customer is told they are talking to an AI, in every channel and every entry point, not only in the terms of service |
| Traceability | Conversations and incidents can be reconstructed and exported in a form an auditor accepts |
| Human oversight | Escalation to a person works in production, including when the agent is under pressure or failing |
| Prohibited practices | No manipulative or deceptive conversational techniques, including implying the agent is human |
| Data protection (GDPR overlap) | Personal data is not exposed, restated or leaked inside the conversation |
| Annex III high-risk flag | If the agent drifts into credit, employment or essential-service decisions, the heavier 2027 regime is triggered |
“Article 50 is a behavioural obligation. You cannot satisfy a behavioural obligation with a policy document.”
Executives of these companies are already decoding conversations
Frequently asked questions
Article 50 sets transparency duties for AI systems that interact with people: the person must be informed that they are dealing with an AI system, unless it is obvious from the context. For chat, voice and messaging agents this means the disclosure has to be present in the flow the customer actually experiences, in every channel. It has applied since 2 August 2026.