European Accessibility Act + EU AI Act: AI-Powered Products Dual Compliance Guide (2026)

On 28 June 2025, the European Accessibility Act (Directive 2019/882, "EAA") became applicable across EU Member States. For the first time, EU law imposes specific, enforceable accessibility requirements on a broad range of consumer-facing digital products and services — including those that incorporate AI components.

If your product includes a chatbot, a recommendation engine, a voice assistant, an AI-powered search interface, or any form of automated decision support that EU consumers interact with, you are now operating under two simultaneous regulatory regimes. The EAA requires that your AI interface be perceivable, operable, understandable, and robust for users with disabilities. The EU AI Act imposes separate but intersecting obligations: it prohibits AI that exploits disability vulnerabilities, requires human oversight mechanisms, and — when your AI component is high-risk — mandates extensive transparency and technical documentation that must itself be accessible.

This is not a future compliance challenge. The EAA is in force now. The EU AI Act's high-risk provisions become fully enforceable on 2 August 2026. You have approximately 14 months to close dual compliance gaps.

This guide maps every substantive intersection between the EAA and the EU AI Act, explains what each regime requires from AI product developers, and gives you a Python compliance checker and a 25-item checklist for identifying and remediating gaps.

Part 1: The European Accessibility Act — AI Products in Scope

EAA Scope: What Products and Services Are Covered

The EAA covers a defined set of products and services that are placed on the EU market after 28 June 2025 (with transition provisions for existing contracts). The covered categories relevant to AI-powered products include:

Products (Annex I, Section I):

• Computers and operating systems — including embedded AI assistants
• Smartphones and mobile operating systems
• Self-service terminals: ATMs, ticketing machines, check-in kiosks (AI-driven fraud detection, accessibility modes)
• Interactive television equipment (AI-powered content recommendation)
• E-readers

Services (Annex I, Section II):

• Electronic communications services — telecoms chatbots, AI-powered customer service IVR
• Audiovisual media services — AI-generated captions, recommendation algorithms
• Air, bus, rail, and waterway passenger transport services — AI-driven booking systems, real-time journey planning
• Banking services — AI-powered credit scoring interfaces, robo-advisory, chatbot support
• E-commerce services — AI recommendation engines, personalized search, checkout chatbots
• E-books and dedicated reading software

The practical consequence: any AI component that is part of a covered service's user interface must comply with the EAA's accessibility requirements. A banking chatbot must be screen-reader compatible. An e-commerce recommendation engine must present its results in a format that assistive technology can parse. An AI-powered journey planner must be operable without a pointing device.

What "Accessible" Means Under the EAA

The EAA's functional accessibility requirements are set out in Annex I. For digital interfaces, the standard maps closely to WCAG 2.1 Level AA (Web Content Accessibility Guidelines, published by W3C), which the EU has formally incorporated via EN 301 549 v3.2.1, the harmonized European accessibility standard.

WCAG 2.1 AA for AI interfaces means:

Perceivable — AI output must be available in forms that users with sensory disabilities can perceive:

• Text alternatives for non-text AI outputs (generated images, audio descriptions, visual charts from AI analytics)
• Captions and audio descriptions for AI-generated video or voice responses
• Sufficient color contrast for AI-generated UI elements (3:1 minimum for large text, 4.5:1 for normal text)
• Resizable text without loss of AI functionality up to 200%

Operable — AI interfaces must be navigable without mouse input:

• Full keyboard accessibility: AI chatbots, configuration panels, AI-generated forms
• No timing requirements that disabled users cannot meet (AI response timers, session timeouts)
• Skip navigation for AI-heavy pages
• Focus management for dynamically injected AI content (ARIA live regions for streaming AI responses)

Understandable — AI interfaces must communicate clearly:

• Plain language explanations of AI outputs (in lay terms, not technical jargon)
• Error identification for AI-assisted input validation
• Consistent navigation and labeling of AI features across the product

Robust — AI-generated content must be parsable by assistive technologies:

• Valid HTML/ARIA semantics for all AI-rendered components
• Compatible with current assistive technologies (NVDA, JAWS, VoiceOver)
• Dynamic AI content updates via ARIA live regions (aria-live="polite" for non-urgent updates, aria-live="assertive" for critical alerts)

EAA Compliance Obligations for AI Product Developers

Under the EAA, manufacturers (which includes software developers who place products on the EU market) must:

• Design and manufacture products that comply with Annex I accessibility requirements (Art.7)
• Prepare technical documentation demonstrating conformity (Art.7(3))
• Affix CE marking and issue an EU Declaration of Conformity (for products in scope)
• Establish procedures to ensure continued accessibility in production (Art.7(4))
• Provide accessible instructions and information to users (Annex I, §1(d))

Service providers must:

• Provide services in compliance with Annex I, Section II requirements (Art.13)
• Include accessibility information in terms and conditions (Art.13(2)(c))
• Provide feedback mechanisms for users to report accessibility issues (Art.13(2)(d))

Part 2: EU AI Act Provisions That Directly Intersect With Disability

Art.5(1)(b): Prohibited AI — Exploiting Disability Vulnerabilities

Art.5(1)(b) of the EU AI Act prohibits AI systems that exploit any of the following vulnerabilities of a specific group of persons: age, disability, or specific social or economic situation, in a way that distorts behavior and causes or is reasonably likely to cause significant harm.

For developers building AI-powered products that may be used by persons with disabilities, Art.5(1)(b) creates a design constraint that operates independently of the EAA:

• Cognitive accessibility AI: An AI system that detects cognitive impairment indicators (slow response time, repetitive questions) and uses this to suppress information, increase urgency pressure, or redirect the user to higher-priced products would likely be prohibited under Art.5(1)(b).
• Adaptive dark patterns: AI personalization that serves more coercive UX patterns to users exhibiting behavioral indicators of disability is prohibited.
• Predatory targeting: AI that identifies users likely to have intellectual disabilities and targets them with financial products or subscription traps is prohibited.

The prohibition applies regardless of whether the AI is also subject to EAA compliance. A product can be EAA-compliant (technically accessible) while simultaneously violating Art.5(1)(b) (using accessibility to exploit). Both regimes apply.

Art.5(1)(c): Prohibited AI — Social Scoring

Art.5(1)(c) prohibits AI systems that make social scoring of natural persons by public authorities or private entities that leads to detrimental treatment in social contexts outside the data's original collection context.

The disability relevance: AI systems that infer disability status from behavioral or biometric indicators and use this inference as a factor in credit scoring, insurance pricing, employment screening, or access to services risk triggering Art.5(1)(c) — detrimental treatment based on disability inference outside any legitimate medical context.

Art.14: Human Oversight Must Be Accessible

Art.14 of the EU AI Act requires that high-risk AI systems be designed so that human operators can "adequately oversee" the system. Art.14(4) specifies the functional properties of human oversight: the ability to understand capabilities and limitations, to monitor operations, to detect anomalies, and to intervene or stop the system.

The EAA–Art.14 intersection is underexplored in compliance literature: if a high-risk AI system is operated by persons with disabilities (as is commonly the case in employment, education, and public services), the human oversight interface itself must comply with the EAA. A high-risk AI used in disability benefit administration that exposes an oversight dashboard to human reviewers must ensure that dashboard is accessible under WCAG 2.1 AA.

This creates a compound obligation: not just "have human oversight" (EU AI Act) but "have accessible human oversight" (EAA + EU AI Act combined).

Art.50: Transparency Disclosure Must Be Accessible

Art.50(1) requires that providers of AI systems interacting with natural persons ensure those persons are informed they are interacting with an AI system. Art.50(2) requires providers of emotion recognition or biometric categorization systems to inform persons exposed to those systems.

The EAA adds the accessibility dimension: the disclosure itself must be perceivable and understandable by users with visual, auditory, or cognitive disabilities. An AI disclosure that only appears as small grey text fails both the EAA (inadequate contrast, not programmatically determinable) and good practice under Art.50.

Annex III High-Risk Classifications: Disability-Adjacent Categories

Several Annex III high-risk AI categories are particularly relevant when the subjects of AI decisions are persons with disabilities:

Category 1 — Education and vocational training: AI systems that determine access to, assessment of, or outcomes within educational programs. If persons with disabilities participate in education (which, under UNCRPD, they must have equal access to), AI used in that context is likely high-risk.

Category 2 — Employment: AI used in recruitment, selection, promotion, or task allocation. An AI recruiter that processes applications from candidates who disclose disability, or that infers disability from behavioral data, is high-risk and must comply with Art.9–15.

Category 3 — Essential private and public services: AI determining access to welfare benefits (including disability benefits), emergency services, and credit. This category directly covers AI in disability benefit administration.

Category 4 — Law enforcement and migration: AI used in border control and asylum decisions — where applicants may include persons with disabilities — is high-risk.

Part 3: The Dual Compliance Matrix

Part 4: Python EAAIActComplianceChecker

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

class EAAProductCategory(Enum):
    COMPUTER_OS = "computer_os"
    SMARTPHONE = "smartphone"
    SELF_SERVICE_TERMINAL = "self_service_terminal"
    ECOMMERCE_SERVICE = "ecommerce_service"
    BANKING_SERVICE = "banking_service"
    TRANSPORT_SERVICE = "transport_service"
    TELECOM_SERVICE = "telecom_service"
    AUDIOVISUAL_SERVICE = "audiovisual_service"
    EBOOK_SERVICE = "ebook_service"
    NOT_IN_SCOPE = "not_in_scope"

class AIActRiskCategory(Enum):
    PROHIBITED = "prohibited"
    HIGH_RISK = "high_risk"
    LIMITED_RISK = "limited_risk"
    MINIMAL_RISK = "minimal_risk"

@dataclass
class AIComponent:
    name: str
    description: str
    interacts_with_users: bool
    uses_biometric_data: bool
    makes_decisions_affecting_disabled: bool
    targets_vulnerable_groups: bool
    has_human_oversight_interface: bool
    oversight_interface_wcag_compliant: Optional[bool] = None
    annex_iii_category: Optional[str] = None  # e.g., "education", "employment", "benefits"

@dataclass
class EAAProduct:
    product_name: str
    eaa_category: EAAProductCategory
    ai_components: list[AIComponent]
    wcag_level_achieved: Optional[str] = None  # "A", "AA", "AAA", None
    has_accessibility_statement: bool = False
    has_feedback_mechanism: bool = False
    has_technical_documentation: bool = False
    placed_on_eu_market_after_june_2025: bool = True

@dataclass
class ComplianceViolation:
    severity: str  # "CRITICAL", "HIGH", "MEDIUM"
    regulation: str  # "EAA", "EU_AI_ACT", "BOTH"
    article: str
    description: str
    remediation: str

@dataclass
class EAAIActComplianceResult:
    product_name: str
    eaa_in_scope: bool
    ai_act_risk_level: AIActRiskCategory
    violations: list[ComplianceViolation] = field(default_factory=list)
    warnings: list[str] = field(default_factory=list)
    compliant: bool = False

class EAAIActComplianceChecker:

    EAA_SCOPED_CATEGORIES = {
        EAAProductCategory.COMPUTER_OS,
        EAAProductCategory.SMARTPHONE,
        EAAProductCategory.SELF_SERVICE_TERMINAL,
        EAAProductCategory.ECOMMERCE_SERVICE,
        EAAProductCategory.BANKING_SERVICE,
        EAAProductCategory.TRANSPORT_SERVICE,
        EAAProductCategory.TELECOM_SERVICE,
        EAAProductCategory.AUDIOVISUAL_SERVICE,
        EAAProductCategory.EBOOK_SERVICE,
    }

    HIGH_RISK_ANNEX_III = {"education", "employment", "benefits", "law_enforcement", "migration", "credit"}

    def assess_dual_compliance(self, product: EAAProduct) -> EAAIActComplianceResult:
        result = EAAIActComplianceResult(
            product_name=product.product_name,
            eaa_in_scope=product.eaa_category in self.EAA_SCOPED_CATEGORIES,
            ai_act_risk_level=self._determine_risk_level(product),
        )

        if result.eaa_in_scope and product.placed_on_eu_market_after_june_2025:
            self._check_eaa_violations(product, result)

        self._check_ai_act_violations(product, result)
        self._check_compound_obligations(product, result)

        result.compliant = len(result.violations) == 0
        result.violations.sort(key=lambda v: ["CRITICAL", "HIGH", "MEDIUM"].index(v.severity))
        return result

    def _determine_risk_level(self, product: EAAProduct) -> AIActRiskCategory:
        for component in product.ai_components:
            if component.targets_vulnerable_groups:
                return AIActRiskCategory.PROHIBITED
            if component.annex_iii_category in self.HIGH_RISK_ANNEX_III:
                return AIActRiskCategory.HIGH_RISK
            if component.interacts_with_users:
                return AIActRiskCategory.LIMITED_RISK
        return AIActRiskCategory.MINIMAL_RISK

    def _check_eaa_violations(self, product: EAAProduct, result: EAAIActComplianceResult):
        if product.wcag_level_achieved not in ("AA", "AAA"):
            result.violations.append(ComplianceViolation(
                severity="CRITICAL",
                regulation="EAA",
                article="Annex I, §1 + EN 301 549 v3.2.1",
                description=f"Product does not meet WCAG 2.1 AA (current: {product.wcag_level_achieved or 'unknown'})",
                remediation="Conduct WCAG 2.1 AA audit. Fix perceivability, operability, understandability, robustness gaps. Re-test with assistive technologies (NVDA/JAWS/VoiceOver).",
            ))
        if not product.has_accessibility_statement:
            result.violations.append(ComplianceViolation(
                severity="HIGH",
                regulation="EAA",
                article="Art.7(3) + Art.13(2)(c)",
                description="No EAA accessibility statement published",
                remediation="Publish EAA accessibility statement documenting conformity status, known limitations, and contact information.",
            ))
        if not product.has_feedback_mechanism:
            result.violations.append(ComplianceViolation(
                severity="HIGH",
                regulation="EAA",
                article="Art.13(2)(d)",
                description="No feedback mechanism for users to report accessibility issues",
                remediation="Implement accessible feedback form or email address. Must itself be accessible (WCAG 2.1 AA).",
            ))
        if not product.has_technical_documentation:
            result.violations.append(ComplianceViolation(
                severity="MEDIUM",
                regulation="EAA",
                article="Art.7(3)",
                description="Technical documentation demonstrating EAA conformity not prepared",
                remediation="Prepare EAA technical documentation: accessible design features, test results, conformity assessment method.",
            ))

    def _check_ai_act_violations(self, product: EAAProduct, result: EAAIActComplianceResult):
        for component in product.ai_components:
            if component.targets_vulnerable_groups:
                result.violations.append(ComplianceViolation(
                    severity="CRITICAL",
                    regulation="EU_AI_ACT",
                    article="Art.5(1)(b)",
                    description=f"AI component '{component.name}' targets vulnerable groups — likely prohibited AI",
                    remediation="Redesign AI component to remove exploitation of disability or other vulnerability vectors. Consult Art.5(1)(b) legal analysis before market placement.",
                ))
            if component.uses_biometric_data and component.makes_decisions_affecting_disabled:
                result.violations.append(ComplianceViolation(
                    severity="HIGH",
                    regulation="EU_AI_ACT",
                    article="Art.5(1)(c) + Art.50(2)",
                    description=f"AI component '{component.name}' uses biometric inference affecting disabled persons — social scoring and transparency risks",
                    remediation="Assess Art.5(1)(c) social scoring risk. Implement Art.50(2) disclosure. Ensure disclosure is WCAG 2.1 AA compliant.",
                ))
            if component.interacts_with_users:
                result.violations.append(ComplianceViolation(
                    severity="MEDIUM",
                    regulation="EU_AI_ACT",
                    article="Art.50(1)",
                    description=f"AI component '{component.name}' interacts with users — Art.50 transparency disclosure required",
                    remediation="Implement Art.50(1) disclosure. Ensure disclosure is accessible: WCAG 2.1 AA contrast, screen-reader compatible, plain language.",
                ))

    def _check_compound_obligations(self, product: EAAProduct, result: EAAIActComplianceResult):
        if result.ai_act_risk_level == AIActRiskCategory.HIGH_RISK and result.eaa_in_scope:
            for component in product.ai_components:
                if component.has_human_oversight_interface and not component.oversight_interface_wcag_compliant:
                    result.violations.append(ComplianceViolation(
                        severity="HIGH",
                        regulation="BOTH",
                        article="EU AI Act Art.14 + EAA Annex I",
                        description=f"Human oversight interface for '{component.name}' not WCAG 2.1 AA compliant — compound violation",
                        remediation="Audit oversight dashboard for WCAG 2.1 AA. High-risk AI oversight must be accessible to disabled operators. Fix keyboard navigation, screen reader labels, contrast.",
                    ))
                if not component.has_human_oversight_interface and result.ai_act_risk_level == AIActRiskCategory.HIGH_RISK:
                    result.violations.append(ComplianceViolation(
                        severity="CRITICAL",
                        regulation="EU_AI_ACT",
                        article="Art.14(4)",
                        description=f"High-risk AI component '{component.name}' has no human oversight interface",
                        remediation="Implement human oversight interface. Must allow: understanding of AI limitations, monitoring, anomaly detection, intervention/stop capability. Must be WCAG 2.1 AA if EAA applies.",
                    ))

# Usage example
checker = EAAIActComplianceChecker()

banking_chatbot = EAAProduct(
    product_name="SmartBank AI Customer Service",
    eaa_category=EAAProductCategory.BANKING_SERVICE,
    wcag_level_achieved="A",  # fails AA
    has_accessibility_statement=False,
    has_feedback_mechanism=True,
    has_technical_documentation=False,
    placed_on_eu_market_after_june_2025=True,
    ai_components=[
        AIComponent(
            name="Customer Service Chatbot",
            description="AI chatbot handling banking queries and support requests",
            interacts_with_users=True,
            uses_biometric_data=False,
            makes_decisions_affecting_disabled=False,
            targets_vulnerable_groups=False,
            has_human_oversight_interface=False,
            annex_iii_category=None,
        ),
        AIComponent(
            name="Credit Scoring Assistant",
            description="AI-powered credit recommendation interface",
            interacts_with_users=True,
            uses_biometric_data=False,
            makes_decisions_affecting_disabled=True,
            targets_vulnerable_groups=False,
            has_human_oversight_interface=True,
            oversight_interface_wcag_compliant=False,
            annex_iii_category="credit",
        ),
    ],
)

result = checker.assess_dual_compliance(banking_chatbot)
print(f"EAA in scope: {result.eaa_in_scope}")
print(f"AI Act risk: {result.ai_act_risk_level.value}")
print(f"Compliant: {result.compliant}")
print(f"\nViolations ({len(result.violations)}):")
for v in result.violations:
    print(f"  [{v.severity}] {v.regulation} {v.article}: {v.description}")

Part 5: 25-Item Dual Compliance Checklist

Part A — EAA Accessibility for AI Interfaces (8 items)

• A1. AI-generated content has text alternatives (WCAG 1.1.1 — Level A)
• A2. AI chatbot/voice interface provides captions or transcripts (WCAG 1.2 — Level AA for live content)
• A3. AI-rendered UI elements meet 4.5:1 contrast ratio (WCAG 1.4.3 — Level AA)
• A4. Full keyboard access to all AI features — no mouse required (WCAG 2.1.1 — Level A)
• A5. AI streaming responses announced via ARIA live regions (aria-live="polite") (WCAG 4.1.3 — Level AA)
• A6. AI input forms have programmatically associated labels, error messages (WCAG 1.3.1, 3.3.1 — Level A)
• A7. EAA accessibility statement published (Art.7(3) / Art.13(2)(c))
• A8. Feedback mechanism for reporting accessibility issues (Art.13(2)(d))

Part B — EU AI Act Prohibited AI (4 items)

• B1. AI does not exploit disability indicators to manipulate behavior (Art.5(1)(b) — prohibited)
• B2. AI does not infer disability status for social scoring outside medical context (Art.5(1)(c))
• B3. AI targeting and personalization logic reviewed for disability exploitation vectors
• B4. Legal Art.5(1)(b) opinion obtained if product serves vulnerable population segments

Part C — Human Oversight Accessibility (4 items)

• C1. Human oversight interface (operator dashboard) is WCAG 2.1 AA compliant (Art.14 + EAA)
• C2. Oversight interface keyboard navigable, screen-reader compatible
• C3. Monitoring and intervention functions accessible without pointing device
• C4. Accessible user manual for oversight interface operations

Part D — Transparency Accessibility (5 items)

• D1. Art.50(1) AI disclosure present at interaction start
• D2. AI disclosure meets WCAG 2.1 AA contrast and text size (≥ 14pt / 18px)
• D3. AI disclosure is programmatically determinable (screen reader announces it)
• D4. AI disclosure in plain language (Flesch-Kincaid Grade 8 or lower)
• D5. Biometric/emotion AI disclosure (Art.50(2)) meets accessibility requirements

Part E — Documentation and Support (4 items)

• E1. Technical documentation (EAA Art.7(3)) prepared and up to date
• E2. User documentation accessible (PDF/A with tagged structure, or accessible HTML)
• E3. Customer support channel accessible to users with hearing or speech impairments
• E4. Conformity assessment process documented (CE marking / EU Declaration of Conformity where applicable)

Part 6: Compliance Timeline

The overlap window — 28 June 2025 to 2 August 2026 — is the dual compliance preparation period. Products that are both EAA-scoped and EU AI Act high-risk have 14 months to achieve dual conformity before the highest-consequence enforcement period begins.

Part 7: Common Implementation Mistakes

Mistake 1: Treating EAA and EU AI Act as Sequential (Do One, Then the Other)

The two regimes have different effective dates but overlapping subject matter. An AI product that achieves WCAG 2.1 AA compliance (EAA) but uses dark-pattern AI personalization that targets disability indicators (Art.5(1)(b)) violates the EU AI Act regardless of EAA compliance. Both regimes must be designed for simultaneously.

Mistake 2: Only Auditing User-Facing Interfaces

The EAA applies to user-facing AI interfaces. But Art.14 of the EU AI Act requires accessible human oversight interfaces — the dashboards and control panels used by operators, not just end users. Many compliance programs audit the end-user product for WCAG compliance but ignore operator-facing AI management tools.

Mistake 3: Generic ARIA Implementation for AI Dynamic Content

AI-generated content — streaming responses, dynamically injected recommendations, real-time classification outputs — requires specific ARIA implementation. A generic aria-live="assertive" on a container that updates every 2 seconds creates an unusable experience for screen reader users (constant interruptions). The correct implementation uses aria-live="polite" for non-urgent updates and manages ARIA region updates carefully. Many AI products that claim WCAG 2.1 AA compliance fail this specific criterion when tested with NVDA or VoiceOver.

Mistake 4: Ignoring EAA for B2B Products

The EAA primarily targets consumer-facing products and services. However, the Art.14 human oversight requirement under the EU AI Act applies to operators — which may include employees with disabilities, public sector workers, or contractors. Even if your AI product is B2B and not EAA-scoped directly, accessible human oversight is best practice and increasingly expected in procurement requirements.

Part 8: Deployment Infrastructure Considerations

EAA and EU AI Act compliance creates specific infrastructure requirements that are easiest to satisfy in an EU-sovereign hosting environment:

Data minimization for accessibility data: WCAG compliance testing and accessibility audit results may contain personal data (test participant data, assistive technology usage patterns). Under GDPR, this data must be processed within the EU or under adequate transfer safeguards. EU-sovereign hosting by design eliminates transfer risk.

Art.14 oversight logging: Human oversight under Art.14 requires maintaining logs of oversight events (when operators intervened, when anomalies were detected). These logs are operational data subject to GDPR. Hosting on EU infrastructure subject to EU law only — not subject to CLOUD Act extraterritorial access — is the lowest-risk architecture for Art.14 logging compliance.

Audit trail accessibility: If your Art.14 oversight system stores structured logs, those logs must be accessible to human reviewers with assistive technologies. EU-sovereign infrastructure running on servers subject only to EU regulation eliminates the risk of third-country government access that could compromise the integrity of oversight records.

sota.io provides EU-sovereign PaaS infrastructure — deploy your AI-powered, EAA-compliant applications on servers located in the EU, operated under EU law, with no CLOUD Act or Patriot Act exposure. GDPR-compliant by architecture, not by policy.

Summary

The European Accessibility Act is not a future concern — it became applicable on 28 June 2025. If your AI-powered product is in scope (e-commerce, banking, transport, telecoms, audiovisual, or consumer devices), you are now under active accessibility obligations.

The EU AI Act adds a parallel layer: prohibited AI that exploits disability (Art.5(1)(b)+(c)), mandatory accessible human oversight (Art.14 + EAA combined), and accessible transparency disclosures (Art.50 + EAA). High-risk AI systems that serve disabled populations carry both regimes simultaneously from August 2026.

The four non-obvious intersections most developers miss:

1. Art.14 human oversight must be WCAG 2.1 AA compliant — not just technically functional
2. Art.50 transparency disclosures must meet EAA accessibility standards — not just be present
3. AI personalization using disability indicators is prohibited — WCAG compliance does not immunize against Art.5(1)(b)
4. EAA applies to B2B AI products via the Art.14 compound obligation — the operator interface must be accessible

Use the Python EAAIActComplianceChecker and 25-item checklist above to identify and prioritize your compliance gaps. The 14-month window to August 2026 is sufficient to close most dual compliance gaps if you start now.