Article 91 — Privileged communications (EU AI Act)

Article 91 of Regulation (EU) 2024/1689 — Privileged communications. Official text, practical interpretation, key obligations and compliance implications.

Official Text Summary

Article 91 of Regulation (EU) 2024/1689 (the EU AI Act) addresses the protection of privileged communications arising in the context of enforcement and market surveillance activities conducted under the Regulation. Positioned within Title IX, which governs post-market monitoring and market surveillance, Article 91 establishes that competent authorities — including market surveillance authorities and the AI Office — must respect the rules on legal professional privilege and equivalent confidentiality protections when receiving, handling, and sharing information from natural or legal persons subject to investigation or inquiry.

The article requires that any information communicated in circumstances that attract legal privilege under applicable Union or national law cannot be used or further disclosed in ways that would undermine the rights of defence of the person concerned. This protection applies irrespective of the format of the communication — whether written, oral, or electronic — provided the conditions for privilege are met under the relevant legal framework.

Article 91 does not create an absolute bar on authority access to information; rather, it imposes a procedural obligation on competent authorities to identify and segregate privileged material, to refrain from relying on such material in enforcement decisions without proper legal basis, and to ensure that individuals and entities retain access to legal counsel throughout the investigative process. This reflects the broader principle, common to EU regulatory enforcement, that procedural rights must be respected even in technically complex or AI-specific enforcement contexts.

What This Means in Practice

For providers, deployers, importers, and distributors of AI systems, Article 91 provides a meaningful procedural safeguard during any investigation, audit, or enforcement action initiated by a national market surveillance authority or the AI Office. In practice, this means that communications between an operator and its legal counsel — including internal legal teams acting in a lawyerly capacity where national law so permits — cannot be compelled for use as evidence against the operator if those communications qualify as legally privileged.

For legal and compliance teams, the immediate implication is that privilege designations should be applied carefully and consistently from the outset of any regulatory interaction. Legal teams advising on AI Act compliance should ensure that advice documents, investigation response strategies, and communications prepared in anticipation of enforcement are properly marked and structured to attract privilege protection under the applicable national law of the Member State where the authority operates.

For competent authorities, Article 91 creates an obligation to implement internal procedures for identifying potentially privileged material received during inspections or information requests. Authorities must avoid using such material in enforcement decisions unless a competent court or authority has determined that privilege does not apply or has been waived.

Concrete example: If a market surveillance authority in Germany conducts an on-site inspection of a provider of a high-risk AI system and seizes documents including emails between the provider's in-house counsel and the engineering team discussing regulatory risk, Article 91 requires the authority to assess whether those emails attract legal professional privilege before relying on them in any enforcement decision or sharing them with other authorities under the Union-level information exchange mechanisms established elsewhere in the Regulation.

Key Obligations

• Competent authorities must respect legal professional privilege and equivalent confidentiality protections when receiving or handling information from persons subject to investigation or enforcement under the EU AI Act.
• Privileged communications may not be used as evidence in enforcement proceedings or decisions without a proper legal basis established under applicable Union or national law.
• Authorities must implement internal segregation procedures to identify and isolate potentially privileged material obtained during inspections, information requests, or other investigative actions.
• Cross-border information sharing must preserve privilege status — when authorities exchange information under the cooperation mechanisms of Title IX and Title X, the privileged character of communications must be communicated and respected by receiving authorities.
• Operators retain the right to legal counsel throughout all stages of investigation and must not be penalised for asserting privilege over communications that lawfully attract such protection.
• The scope of privilege is determined by applicable national law in the Member State of the authority concerned, subject to any Union-level minimum standards on the rights of defence.

Relationship to Other Articles

Article 91 operates within the broader enforcement and market surveillance architecture of the EU AI Act and should be read in conjunction with several other provisions.

Articles 74 to 83 (Title IX) establish the general framework for market surveillance, including the powers of national authorities to conduct inspections, request information, and take corrective measures — all of which are subject to the procedural constraints imposed by Article 91.

Article 88 (access to data and documentation) and Article 90 (reporting of serious incidents) create specific information disclosure obligations that must be exercised consistently with the privilege protections in Article 91; an obligation to disclose does not automatically override applicable privilege, and competent authorities must navigate this tension under national procedural law.

Articles 70 and 71 address confidentiality and information sharing between authorities at Union level, including through the AI Office and the European Artificial Intelligence Board. The confidentiality obligations in those articles interact with Article 91 by establishing baseline protections for competitively sensitive information, to which Article 91 adds the distinct layer of legal professional privilege.

Recitals 148 to 152 provide interpretive context for the enforcement architecture, including the principle that rights of defence must be respected throughout AI Act proceedings.

Compliance Timeline

2 August 2024 — Regulation (EU) 2024/1689 entered into force. Article 91, as a procedural provision within Title IX, became part of the binding legal framework from this date, though enforcement activity under the full surveillance regime had not yet commenced.

2 February 2025 — Prohibition on unacceptable-risk AI practices became applicable (Article 5). Any enforcement actions taken in relation to prohibited AI systems from this date are subject to the procedural protections in Article 91.

2 August 2025 — GPAI model obligations became applicable. Enforcement activity targeting general-purpose AI model providers is also governed by Article 91 from this date.

2 August 2026 — The AI Office and national authorities are expected to be fully operational with respect to market surveillance of high-risk AI systems under Annex I. Privilege protections become operationally significant as enforcement caseloads increase.

2 August 2027 — Full application of obligations for the remaining categories of high-risk AI systems (Annex III). From this date, the complete market surveillance regime is active and Article 91 applies across all enforcement contexts contemplated by the Regulation.

Operators engaged in any regulatory dialogue with competent authorities should treat Article 91 protections as immediately operative whenever a formal or informal investigation is initiated, regardless of the specific phased date applicable to the AI system category under review.