Artificial Intelligence Policy Brief

16.3.2025

Executive Summary

Artificial intelligence (AI) is rapidly reshaping society, and autism has emerged as a prominent focus of AI research and development. This policy brief is a culmination of two years of research as part of Project AIRA (Artificial Intelligence and the Rights of Autistic People), including a literature review of scientific studies on the intersections of autism and AI, and a survey of autistic people across Europe on their perspectives regarding artificial intelligence. Our research found that while autistic people are aware of AI’s potential benefits, particularly for aiding screening, diagnosis and supporting everyday living, they express serious concerns about consent and privacy, algorithmic bias, and AI systems perpetuating discrimination. A clear majority (90%)  emphasized the vital importance of autistic involvement in the development of AI applications that target autism, even as they remain largely excluded from research and decision-making processes.

We propose a framework of comprehensive protections and guidelines that embed meaningful autistic participation throughout the AI development lifecycle. This includes requirements for representative datasets to prevent discriminatory impacts, a shift toward AI applications that enhance autonomy rather than modify behavior, and robust safeguards to ensure informed consent regarding use of sensitive genetic, biomedical and cognition data. We call for prioritizing funding for projects that demonstrate authentic collaboration with autistic communities and embrace neurodiversity-affirming principles, ensuring the development of AI applications that truly serve autistic people’s needs while protecting against potential harms.

Need for Action: Protecting the Rights of Autistic People in the AI Era

From exclusion to involvement in Research and Development

A troubling trend dominates the rapid growth in applying AI in autism research: AI development is being done to autistic people, not with us. While the majority (90%) of autistic people surveyed in our project wanted to be involved in developing AI applications that affect their lives, but they are systematically excluded from research design and policy decisions. We observed the same trend in our study that mapped over 1,000 autism research studies involving AI, highlighting how current research priorities actively conflict with the needs of the autistic communities.

  • Early Detection & Diagnosis: 80% of the studies we mapped deployed AI for early detection, primarily in young children. While providing early diagnosis undeniably opens up the possibility of providing vital support, this imbalance neglects the pressing needs of autistic adults who struggle to access diagnostic services due to cost barriers, specialist shortages, and institutional indifference—a critical oversight that demands attention to bring the benefits of AI more equally within reach of all autistic people.
  • Biomedical Research: Genetic, epigenetic and neurological causes and related treatments have been historically emphasized in autism research, perpetuating a deficit-based medical model, despite growing recognition that autism represents natural human diversity rather than a condition requiring correction. While autistic communities welcome some types of biomedical research, the extreme emphasis of applying  AI in these types of research, and neglecting potential uses in other fields, risks amplifying pathologizing perspectives
  • Behavioral Interventions: While autistic people and their advocacy organizations have strongly objected to behavioral interventions, AI has been applied to intensify these types of interventions, in stark contrast to the absence of similar work on support strategies that autistic experts and autistic advocacy groups recommend.
  • Priorities identified by autistic communities – such as assessment of support needs, improving delivery of support services, communication support, practical everyday living support and large scale monitoring of human rights violations receive less than 10% of research attention.

In sum, despite substantial investment in developing AI-based screening and diagnostic tools, the pressing needs of autistic adults without access to affordable diagnostic services continue to be systematically overlooked. Thus, AI is not being used in ways that could address a critical gap in support systems. This failure to meaningfully incorporate autistic perspectives has resulted in AI solutions that not only fail to address real needs but potentially inflict additional harm on the very individuals they purport to serve.

Addressing Heightened Vulnerability to AI-Enabled Manipulation

Our research identifies four critical vulnerabilities that put autistic people at particular risk from AI systems:

  1. Forced Conformity: AI can be used to intensify behavioral interventions that pressure autistic people to suppress their natural ways of being, and instead perform neurotypical behaviors – a practice called ‘masking’. Cognitive manipulation that leads to masking restricts authentic self-expression and undermines personal autonomy and free will.
  2. Biometric Misinterpretation Risks: There is a risk that AI systems trained on datasets of predominantly “normal” behavior and expression may misclassify natural autistic expression as “abnormal”, suspicious or otherwise undesirable, leading to automated discrimination. This creates particularly acute vulnerabilities for autistic people with intersecting marginalized identities, such as autistic refugees, who face compounded discrimination when their neurological differences are flagged as suspicious.
  3. Privacy and Consent Concerns: True informed consent requires autistic individuals to fully understand how their personal data will be used, shared between studies, and distributed globally. The introduction of AI and its application to large international and global datasets is making this increasingly difficult, creating the risk that autistic people are manipulated to support autism research they do not actually approve of. Further, 80% of the participants in our survey had major concerns about data security, highlighting gaps in ethical standards and privacy protections – particularly for biological data from autistic children.

Impact of Inaction

Time is running out to establish proper safeguards as AI rapidly expands into employment, healthcare, and education, and regulatory guidelines become entrenched. Without immediate policy action, discriminatory practices risk becoming permanent fixtures in AI systems, with long-term effects. We call for an urgent and fundamental shift in how autism-focused AI is researched, developed, and regulated.

Proposed Policy Option(s)

Two fundamental policy interventions must be implemented to ensure AI development serves rather than harms autistic people. These strategic options are complementary and mutually reinforcing. While the first ensures AI development aligns with community needs and values, the second provides essential protections against potential harms. Together, they form a comprehensive approach to prevent the marginalization of autistic people as a result of AI development.

1: Meaningful Inclusion in AI Development

Autistic people must co-lead AI development that affects them, not just be its subjects – reflecting the broader shift from viewing autism as something to “cure” to recognizing it as part of human diversity, involving both beneficial characteristics and disability. Token inclusion is not enough: Research teams must actively partner with autistic communities rather than waiting for demands for representation. To support this change, funding bodies should prioritize projects that implement community-proposed solutions through genuine collaboration between researchers and autistic people’s advocacy organisations.

This transformation requires fundamentally changing research practices, shifting priorities to address the actual needs of autistic people rather than assumed deficits. Resources should focus on AI tools that enhance wellbeing and autonomy instead of enforcing superficial conformity, while ensuring systems are built with datasets reflecting full human diversity, thoroughly tested for bias. Further, we recommend establishing training and mentorship programs led by autistic advocacy experts and AI professionals, supporting autistic individuals in decision-making roles across research and development settings. Expanding AI research into Augmentative and Alternative Communication (AAC) tools is also essential to broaden the range of autistic people with access to genuine decision-making.

2: Strengthened EU Regulatory Framework

EU AI regulations need stronger protections against automated discrimination of autistic people. While the AI Office develops guidelines for the AI Act, it must proactively engage with autistic people’s advocacy organizations, especially regarding medium and high-risk systems – placing responsibility for consultation on decision-makers rather than resource-limited autistic communities. The framework must address core weaknesses in current guidelines, particularly examining exemptions in healthcare, autism interventions, research, and national security that could enable discrimination despite existing protections. For AI systems using autistic people’s data, comprehensive protections must include accessible informed consent practices, transparent data collection, and strong safeguards against biometric surveillance.

Policy Recommendations

1. Formal Involvement of Autism and Neurodiversity Expertise

  • Create an Autism Advisory Panel within the European AI Board to evaluate medium and high-risk systems and issue guidelines from the perspective of autism and neuro- affirming principles. The panel should have authority to request audits when systems fail to meet standards, and participate in conformity assessments before market approval. Working closely with national authorities implementing the AI Act, the panel would ensure considerations for the rights of autistic people are embedded in AI governance across Europe. The panel should include representatives from autistic people’s advocacy organizations and AI ethics experts to ensure comprehensive expertise.

2. Prioritize Inclusive and Empowering Innovation:

  • Realign Research Priorities: The European Research Council (ERC) and Horizon Europe should establish minimum funding thresholds for AI projects addressing priorities identified by autistic people’s advocacy organisations, such as communication support, tools to support independent living, neuro-affirming interventions, collecting and monitoring data on realization of human rights and welfare services, accessible screening and diagnostics for autistic people of all ages. These thresholds should be established in consultation with autistic people’s organisations, with initial consultations beginning within the next-possible funding cycle.
  • Mandate Authentic Collaboration: Require EU-funded projects that apply AI in autism research to demonstrate meaningful engagement with autistic people’s organisations and autistic experts throughout the entire research lifecycle, from initial design to implementation and evaluation. Multiple pathways for fulfilling this requirement should be provided based on project scope and available resources.

3. Enhanced Engagement of the EU AI Office with Autism Advocacy Groups

  • The EU AI Act presents a crucial opportunity to strengthen autistic people’s rights. As the EU AI Office develops guidance for implementation from 2025-2027, it must establish formal consultation with autistic people’s organizations. This engagement will ensure regulations effectively protect autistic individuals while creating a model that balances innovation with fundamental rights.
  • Key Articles Requiring Attention:
    • Article 2(8): Address the research exemption of the Act by establish clear oversight criteria for high-risk AI research involving autistic populations, including mandatory ethics review, data protection standards, and regular progress monitoring.
    • Article 5(1)(a): Explore ways to include autism and other forms of neurodivergence as specific vulnerability factors when assessing manipulative AI systems.
    • Article 5(1)(b): Establish clear criteria for assessing exploitation risk in AI systems targeting autistic populations.
    • Article 10: Introduce minimum diversity thresholds for autism-related training data, requiring proportional representation across gender, cultural backgrounds, and support needs.
    • Article 16: Require providers of high-risk educational and employment AI systems to demonstrate substantive engagement with autistic and other neurodivergent stakeholders.

4. Establish Comprehensive Data Protection Standards

  • Strengthen Informed Consent Requirements: Mandate accessible informed consent processes for AI systems collecting or processing autistic people’s data, with particular protections for biological, genetic, and behavioral data collected from autistic children and vulnerable adults.
  • Improved Data Governance Framework: Develop transparent guidelines for how personal data can be used, shared across research initiatives, and distributed through international or global databases, with clear opt-out mechanisms and regular audits of compliance.

EUCAP. (2021). Autistic priorities survey 2021. European Council of Autistic People. http://eucap.eu/projects/autistic-priorities/
EUCAP. (2022). Survey on autistic people’s experiences with ABA. European Council of Autistic People. http://eucap.eu/projects/aba/
EUCAP. (2024). Autistic people’s views on ABA in Europe: EUCAP survey results 2022. European Council of Autistic People. http://eucap.eu/projects/aba
Global Autistic Task Force on Autism Research (GATFAR). (2022). Open letter to the Lancet Commission. http://eucap.eu/2022/02/14/open-letter-to-lancet-commission/
EUCAP & AIRA Project Group. (2024). Survey report: Artificial intelligence and the rights of autistic people (AIRA). European Council of Autistic People. http://eucap.eu/ai-advocacy/