EU AI Act Art.63: Advisory Forum — Multi-Stakeholder Consultation, Composition, Tasks, and Role in AI Governance (2026)

EU AI Act Article 63 establishes the Advisory Forum — the multi-stakeholder consultation body through which industry representatives, civil society organisations, academic researchers, SMEs, and standards bodies provide formal input into AI Act implementation. Where the AI Board (Art.59) coordinates national competent authorities and the Scientific Panel (Art.61) provides independent technical expert opinions for enforcement, the Advisory Forum is the stakeholder voice layer of the Chapter VI governance architecture.

The Advisory Forum is not an enforcement body. It cannot impose obligations, issue binding opinions, or compel any regulatory action by the AI Office or AI Board. Its institutional function is legitimacy and practical grounding: by systematically incorporating perspectives from those who build, deploy, study, and are affected by AI systems, the Advisory Forum ensures that the AI Office's implementing measures, guidance documents, and codes-of-practice assessments reflect operational reality rather than purely regulatory abstraction.

For developers and providers of AI systems — and for EU-sovereign infrastructure providers whose platforms host AI workloads — Art.63 defines the formal channel through which you can participate in shaping the rules that govern your products. Understanding the Advisory Forum's composition requirements, tasks, and procedural constraints is prerequisite to effective engagement.

Art.63 became applicable on 2 August 2025 as part of the phased entry into force of Regulation (EU) 2024/1689.

Art.63 in the Chapter VI Governance Architecture

The Advisory Forum occupies a distinct position in the Chapter VI governance framework, deliberately separated from both the political-coordination tier (AI Board) and the independent-expertise tier (Scientific Panel):

The critical structural distinction between the Advisory Forum and Scientific Panel:

• Advisory Forum: Includes industry representatives with direct commercial interests → provides practical implementation perspectives, can advocate for stakeholder positions, but cannot participate in enforcement decisions where conflict of interest exists
• Scientific Panel: No industry representatives → independent technical evaluation for enforcement proceedings, conflict-of-interest declarations required, provides the technical evidentiary foundation for AI Office corrective measures

These bodies are complementary, not duplicative. The Advisory Forum shapes how rules are written and implemented; the Scientific Panel evaluates whether specific AI systems comply with those rules.

Art.63(1): Commission Mandate to Establish the Advisory Forum

Art.63(1) places the establishment obligation on the European Commission: the Commission shall establish an Advisory Forum to provide technical expertise and advice to the AI Board and the AI Office on the implementation of the Regulation.

Scope of advice: The Advisory Forum's mandate covers the full implementation scope of the AI Act — not merely GPAI model obligations (Chapter V) or high-risk AI requirements (Chapter III), but the Act as a whole. This breadth reflects the Forum's role as a horizontal input mechanism for AI Act implementation rather than a body with a narrow technical remit.

Advice recipients: The Advisory Forum advises both the AI Board (political coordination tier) and the AI Office (enforcement tier). This dual channel prevents the Forum's input from being filtered by either body alone — stakeholder perspectives reach the Commission officials operationalising the Act and the national authority representatives responsible for its implementation simultaneously.

Permanence: The Advisory Forum is a standing body, not an ad hoc expert group. This permanence is operationally significant: ongoing AI Act implementation — updating risk classifications, revising codes of practice, adjusting standardisation mandates, responding to emerging AI capabilities — requires a consultation body with institutional continuity and accumulated implementation knowledge.

Art.63(2): Composition Requirements

Art.63(2) establishes the composition of the Advisory Forum, with explicit requirements designed to prevent any single stakeholder category from dominating.

Stakeholder Categories

The Advisory Forum must include balanced representation from:

Industry — AI providers and deployers:

• Developers and providers of AI systems (including GPAI model providers subject to Chapter V)
• Deployers — organisations that integrate AI systems into products and services
• AI infrastructure providers (cloud platforms, compute infrastructure)
• Industry associations representing AI sector participants

Civil society:

• Non-governmental organisations with AI, digital rights, or fundamental rights mandates
• Consumer protection organisations
• Organisations representing affected communities (disability, labour, vulnerable groups)
• Data protection civil society groups

Academic and research institutions:

• Universities and research centres conducting AI safety, fairness, or policy research
• Standards research bodies
• AI ethics researchers and institutes

Small and medium-sized enterprises:

• SME-specific representation to prevent large-enterprise capture of the forum
• SMEs participate both as AI system developers and as AI system deployers
• AI Act compliance imposes proportionally higher burdens on smaller organisations — their operational perspective is structurally distinct from that of large enterprises

Standards bodies:

• European standards organisations (CEN-CENELEC, ETSI)
• National standards bodies where relevant
• International standards engagement (ISO/IEC JTC 1 liaison)
• Standards bodies provide the bridge between Advisory Forum input and the formal harmonised standards process under Art.40

SME Representation Guarantee

Art.63(2) includes an explicit requirement that a meaningful proportion of the Forum's industry membership consists of SME representatives. This guarantee reflects a deliberate legislative choice: without structural protection, advisory forums for technical regulation systematically skew toward large-enterprise perspectives that can afford professional regulatory engagement teams. The SME guarantee prevents the Advisory Forum from becoming a mechanism through which large incumbents effectively co-author rules that impose barriers to smaller competitors.

Geographic Balance

The Commission is required to ensure geographic balance across EU member states in Forum composition. This requirement prevents the Advisory Forum from being dominated by AI industry actors concentrated in specific member states (notably those with large tech hub concentrations), ensuring that implementation experience from smaller AI ecosystems and more AI-cautious member states informs the Forum's outputs.

Rotating Membership

Forum membership is not permanent. The Commission selects members through an open call process with defined terms, ensuring that the Forum's composition reflects the evolving landscape of EU AI stakeholders rather than becoming entrenched with the organisations that were prominent at the time of the Act's entry into force.

Art.63(3): Tasks of the Advisory Forum

Art.63(3) enumerates the Advisory Forum's mandate. The Forum's tasks span four functional areas: regulatory input, standardisation support, implementation assistance, and reporting.

Regulatory Input and Advisory Opinions

Input to AI Office and AI Board guidance: The Advisory Forum provides technical perspectives that inform the guidance documents, Q&A publications, and implementing acts issued by the AI Office and AI Board. When the AI Office prepares guidance on how to classify AI systems under Chapter III, or how deployers should conduct fundamental rights impact assessments (Art.27), Advisory Forum input grounds the guidance in operational implementation experience.

Codes of practice (Art.56): The Advisory Forum plays a formal role in the Art.56 codes-of-practice development process. GPAI model providers establish codes of practice with the support of the AI Office, and the Advisory Forum provides stakeholder perspectives on:

• Whether proposed CoP technical commitments are operationally feasible for providers of varying sizes
• Whether CoP monitoring and reporting requirements are practically implementable
• Whether CoP transparency obligations adequately serve civil society and downstream deployer information needs
• Whether SME providers can realistically comply with CoP obligations designed primarily with large-model providers in mind

Delegated and implementing acts: When the Commission prepares delegated acts (e.g., updating the high-risk AI system list in Annex III, adjusting the systemic risk threshold in Art.51) or implementing acts (e.g., harmonised templates for technical documentation), the Advisory Forum provides the practical implementation perspective that informs proportionality assessments.

Standardisation Support

Harmonised standards development (Art.40): The Advisory Forum provides input to the standardisation process through its standards body members. CEN-CENELEC and ETSI representatives on the Forum connect the regulatory implementation timeline to the standards development pipeline, flagging gaps between what the Act requires and what harmonised standards currently cover, and prioritising standardisation work based on practical implementation urgency.

Technical specification contribution: Where harmonised standards are absent or incomplete, the Commission can adopt common specifications (Art.41). Advisory Forum input shapes which technical areas need common specifications and what level of technical prescriptiveness is appropriate given industry capabilities.

Annual Review and Reporting

Contribution to annual implementation report: The Commission reports annually to the European Parliament and Council on AI Act implementation. The Advisory Forum provides structured input to this report — including implementation challenges encountered by providers and deployers, observed compliance costs, emerging AI capability categories that may require regulatory attention, and gaps in current guidance or standardisation coverage.

Review of risk classification and prohibitions: As AI capabilities evolve, the Act's prohibited practices (Art.5) and high-risk classifications (Annex III) require updating. Advisory Forum input informs the Commission's assessment of whether current classifications remain calibrated to actual risk, and whether new AI applications require regulatory attention.

Art.63(4): Appointment Process and Terms

Art.63(4) establishes the process through which Advisory Forum members are selected and the terms under which they serve.

Open call: The Commission selects Advisory Forum members through an open, transparent call for expressions of interest. This open process prevents the Forum from being composed exclusively of organisations with pre-existing relationships with Commission services, and enables participation from stakeholder communities that may not have established Brussels presence.

Selection criteria: The Commission selects members based on:

• Demonstrated expertise and practical experience in AI development, deployment, or impact assessment
• Relevance to the stakeholder category they represent (an NGO claiming to represent civil society perspectives must have demonstrable civil society mandate)
• Geographic and sectoral balance requirements
• Independence from other Advisory Forum members in terms of market position (avoiding a Forum dominated by a single large platform company plus its satellite organisations)

Term length: Advisory Forum members serve for two-year renewable terms. The two-year cycle is shorter than the Scientific Panel's three-year appointments, reflecting the Advisory Forum's broader stakeholder composition and the higher rate at which stakeholder interests and organisational capacities evolve in the AI sector.

Renewal: Terms are renewable, allowing organisations with strong track records of constructive engagement to maintain Forum participation while ensuring that the open call process periodically opens access to stakeholders who may not have been selected in prior cycles.

Removal: Members who cease to meet the criteria under which they were selected — for example, an SME representative whose organisation grows beyond the SME threshold — or who breach procedural obligations (confidentiality, conflict-of-interest disclosure) may be removed by the Commission.

Art.63(5): Personal Capacity and Conflict-of-Interest Obligations

Art.63(5) addresses how Advisory Forum members serve and how conflicts of interest are managed.

Personal Capacity vs. Institutional Representation

Advisory Forum members serve in their personal capacity as experts representing a stakeholder category — not as official representatives of the employing organisation or industry association that nominated them. This distinction has practical consequences:

• An engineer from a large AI provider serves the Forum as an AI developer expert, not as the company's representative
• The company cannot direct the engineer's Forum positions as an agent-principal relationship
• Forum outputs reflect the member's expert judgment, not the institutional position of their employer

However, the distinction is less absolute than for the Scientific Panel. Advisory Forum members are selected precisely because they represent stakeholder categories — an industry member is expected to bring industry perspectives, not pretend to be independent of industry interests. The personal-capacity requirement prevents organisations from formally committing to Forum positions without internal deliberation, not from bringing industry perspectives to Forum discussions.

Conflict-of-Interest Declarations

Members are required to declare annually any interests that could affect the impartiality of their contributions to Forum deliberations. The declaration covers:

• Financial interests in AI companies whose products may be affected by Forum outputs
• Professional relationships with organisations under AI Office investigation
• Contractual obligations to industry associations with positions on matters before the Forum
• Research funding from organisations with direct interests in Forum outputs

Recusal mechanism: Where a declared conflict of interest is directly material to a specific Forum discussion or opinion, the affected member is required to recuse from that specific discussion. Recusal does not constitute a finding of impropriety — it is a procedural safeguard that preserves the integrity of the Forum's outputs without disqualifying industry members from participation in the Forum as a whole.

Art.63(6): Rules of Procedure

Art.63(6) requires the Commission to adopt rules of procedure for the Advisory Forum through an implementing act.

The rules of procedure establish the operational mechanics of Forum functioning: meeting frequency, quorum requirements, decision-making processes (consensus vs. majority), working group structure, document classification (public vs. restricted), public consultation procedures, interaction protocols with AI Board and AI Office, and the format for Forum opinions and recommendations.

Public consultation obligations: The Advisory Forum is not itself a public consultation mechanism — it is a selected multi-stakeholder body. However, the Forum's rules of procedure may require the Forum to conduct or reference broader public consultations before finalising positions on significant matters such as codes of practice adequacy or standardisation priorities.

Transparency: Forum meeting agendas, minutes (subject to confidentiality redactions), and formal opinions are typically published on the AI Office website. This publication obligation ensures that Forum outputs are available to the broader AI developer community and not restricted to the Forum membership.

Art.63(7): Commission Secretariat and Support

Art.63(7) provides that the Commission provides the Advisory Forum's secretariat and ensures the material and organisational support necessary for the Forum's functioning.

Secretariat role: The Commission secretariat handles meeting logistics, document preparation and distribution, record-keeping, publication of Forum outputs, and coordination between the Advisory Forum and the AI Board/AI Office. The secretariat does not have a substantive influence on Forum positions but serves as the administrative backbone enabling Forum operations.

Funding: Forum members typically serve without compensation for their participation — they are expected to represent their stakeholder category as part of their professional responsibilities. The Commission covers meeting costs (interpretation, travel for in-person meetings, administrative costs) but not members' opportunity costs.

Independence from commercial influence: The Commission's role as secretariat and funder prevents any single commercial stakeholder from gaining administrative leverage over Forum operations through funding relationships. This structural design distinguishes the Advisory Forum from some national AI advisory councils that depend partly on industry funding and may be susceptible to corresponding influence.

Advisory Forum vs. Scientific Panel: The Critical Distinction for Enforcement

The Advisory Forum and Scientific Panel are complementary, not competing, governance bodies — but their roles diverge sharply at the point of enforcement:

The most significant practical difference: a GPAI model provider can participate in the Advisory Forum — their engineers can contribute to codes-of-practice development, provide input on implementation guidance, and share operational perspectives on GPAI compliance challenges. The same provider's documentation will be evaluated by the Scientific Panel in enforcement proceedings — but the panel members evaluating that documentation are strictly independent of the provider and any organisation with a financial relationship to them.

This two-tier design enables meaningful industry input into AI Act implementation while preserving the integrity of enforcement: you help write the rules through the Advisory Forum, but independent experts assess whether you comply with those rules.

Advisory Forum Role in the Art.56 Codes-of-Practice Process

The Advisory Forum's most operationally significant function for AI providers is its role in the Art.56 codes-of-practice (CoP) development process.

Art.56 requires GPAI model providers with systemic risk (Art.51 designation) to adhere to a code of practice. The AI Office facilitates CoP development with provider participation. The Advisory Forum contributes to this process at three points:

1. Drafting input: Advisory Forum industry members — including GPAI model providers — participate in the CoP drafting process facilitated by the AI Office. Their technical contributions shape the specific obligations, testing methodologies, incident reporting formats, and documentation standards that the CoP embeds.

2. Civil society and SME perspective: Advisory Forum civil society members ensure that CoP obligations address the transparency needs of civil society and the capability constraints of smaller model providers. Without structured civil society input, CoP obligations could be calibrated to the capabilities of large foundation model providers in ways that expose civil society to information asymmetries.

3. Standards body interface: Advisory Forum standards body members connect CoP technical commitments to the formal standardisation pipeline, ensuring that CoP obligations are either supported by existing harmonised standards or that gaps trigger priority standardisation work.

The Advisory Forum does not conduct the assessment of whether a CoP provides adequate compliance assurance — that is the Scientific Panel's function under Art.61(4). The Forum shapes the CoP content; the Panel evaluates its adequacy.

CLOUD Act Implications for Advisory Forum Participation

Multi-national AI companies subject to US jurisdiction face a structural tension when participating in the Advisory Forum:

Disclosure risk: Advisory Forum meetings may involve discussion of other companies' compliance approaches, regulatory interpretation positions, or AI capability assessments. For a US company whose Forum representative could be subject to a US FISA order or CLOUD Act request, participation in Forum discussions involving commercially sensitive third-party information creates a disclosure risk.

EU confidentiality protections: EU law provides confidentiality protections for proprietary business information shared in regulatory proceedings. Advisory Forum rules of procedure typically restrict the use of information shared in Forum contexts. However, these EU-law protections do not directly restrain US legal processes — a US company's employees can be compelled to produce information to US authorities regardless of EU confidentiality obligations.

Practical mitigation: Companies subject to US jurisdiction should:

• Carefully delineate what proprietary competitor information they receive in Advisory Forum contexts versus what is publicly available
• Ensure Forum participation is handled by employees with clear guidance on US legal obligations
• Consider whether participation in restricted-distribution Forum working groups on sensitive topics creates CLOUD Act exposure disproportionate to the participation benefit

Sota.io angle: EU-sovereign infrastructure providers face no structural CLOUD Act exposure from Advisory Forum participation. Their engineering contributions to CoP development and standardisation input can proceed without the dual-jurisdiction tension that constrains US-headquartered participants.

Python Implementation: Advisory Forum Participation Tracker

from dataclasses import dataclass, field
from datetime import date
from enum import Enum
from typing import Optional


class StakeholderCategory(str, Enum):
    PROVIDER = "provider"
    DEPLOYER = "deployer"
    CIVIL_SOCIETY = "civil_society"
    ACADEMIA = "academia"
    SME = "sme"
    STANDARDS_BODY = "standards_body"
    INFRASTRUCTURE = "infrastructure"


class ForumOpinionStatus(str, Enum):
    SUBMITTED = "submitted"
    UNDER_REVIEW = "under_review"
    ADOPTED = "adopted"
    NOT_ADOPTED = "not_adopted"
    SUPERSEDED = "superseded"


@dataclass
class ConflictOfInterestDeclaration:
    declared_date: date
    description: str
    affected_topics: list[str]
    recusal_required: bool

    def requires_recusal_for(self, topic: str) -> bool:
        return self.recusal_required and any(
            t.lower() in topic.lower() for t in self.affected_topics
        )


@dataclass
class AdvisoryForumMember:
    name: str
    organisation: str
    category: StakeholderCategory
    appointment_date: date
    term_end_date: date
    is_sme_representative: bool = False
    member_state: Optional[str] = None
    conflict_declarations: list[ConflictOfInterestDeclaration] = field(
        default_factory=list
    )

    @property
    def is_active(self) -> bool:
        return date.today() <= self.term_end_date

    def check_recusal(self, topic: str) -> bool:
        return any(
            decl.requires_recusal_for(topic) for decl in self.conflict_declarations
        )

    def cloud_act_risk(self) -> str:
        """Assess CLOUD Act exposure risk for US-jurisdiction entities."""
        us_jurisdiction_indicators = ["Inc.", "LLC", "Corp.", "US", "American"]
        if any(ind in self.organisation for ind in us_jurisdiction_indicators):
            return "HIGH — US-incorporated entity; CLOUD Act/FISA exposure for proprietary information shared in restricted Forum sessions"
        return "LOW — EU-incorporated entity; no structural US jurisdiction exposure"


@dataclass
class AdvisoryForumOpinion:
    reference: str
    topic: str
    addressed_to: str  # "AI Office", "AI Board", "Commission"
    submission_date: date
    status: ForumOpinionStatus
    related_article: Optional[str] = None
    cop_contribution: bool = False

    def opinion_summary(self) -> str:
        cop_note = " [CoP contribution]" if self.cop_contribution else ""
        return (
            f"Opinion {self.reference}: {self.topic} → {self.addressed_to} "
            f"({self.status.value}){cop_note} — submitted {self.submission_date}"
        )


class AdvisoryForumTracker:
    def __init__(self):
        self.members: list[AdvisoryForumMember] = []
        self.opinions: list[AdvisoryForumOpinion] = []

    def add_member(self, member: AdvisoryForumMember) -> None:
        self.members.append(member)

    def active_members_by_category(self) -> dict[StakeholderCategory, list[AdvisoryForumMember]]:
        result: dict[StakeholderCategory, list[AdvisoryForumMember]] = {}
        for m in self.members:
            if m.is_active:
                result.setdefault(m.category, []).append(m)
        return result

    def sme_representation_ratio(self) -> float:
        """Art.63(2): SME representation ratio across industry members."""
        industry_cats = {StakeholderCategory.PROVIDER, StakeholderCategory.DEPLOYER, StakeholderCategory.INFRASTRUCTURE}
        industry_members = [m for m in self.members if m.is_active and m.category in industry_cats]
        if not industry_members:
            return 0.0
        sme_members = [m for m in industry_members if m.is_sme_representative]
        return len(sme_members) / len(industry_members)

    def members_eligible_for_topic(self, topic: str) -> list[AdvisoryForumMember]:
        """Returns active members not recused from the given topic."""
        return [
            m for m in self.members
            if m.is_active and not m.check_recusal(topic)
        ]

    def add_opinion(self, opinion: AdvisoryForumOpinion) -> None:
        self.opinions.append(opinion)

    def cop_contributions(self) -> list[AdvisoryForumOpinion]:
        return [o for o in self.opinions if o.cop_contribution]

    def forum_report(self) -> str:
        active = [m for m in self.members if m.is_active]
        by_cat = self.active_members_by_category()
        sme_ratio = self.sme_representation_ratio()
        cop_count = len(self.cop_contributions())
        lines = [
            f"Advisory Forum Status — {date.today()}",
            f"Active members: {len(active)}",
            *(f"  {cat.value}: {len(mems)}" for cat, mems in by_cat.items()),
            f"SME representation (industry members): {sme_ratio:.0%}",
            f"Opinions submitted: {len(self.opinions)}",
            f"CoP contributions: {cop_count}",
        ]
        return "\n".join(lines)

Art.63 Participation Checklist

Series Context: Chapter VI Governance Framework

EU AI Act Art.63 analysis based on Regulation (EU) 2024/1689 as published in the Official Journal of the European Union. Applicable from 2 August 2025 per Art.113(3). Advisory Forum rules of procedure, membership selection criteria, and conflict-of-interest declaration templates will be established through Commission implementing acts. This guide reflects the text of the Regulation as enacted.