M-Enabling Summit 2026

No one is ever accidentally accessible.” IDA at the M-Enabling Summit 2026

At the M-Enabling Summit in Arlington, IDA and G3ict brought together people who build, buy, regulate and live with AI systems.

What happened

This week, IDA did something new: we went to the M-Enabling Summit to start working directly with the private sector on AI. On Wednesday morning, IDA and G3ict convened a roundtable where participants worked through real examples of where AI systems still fail persons with disabilities. They then began setting out what a governance standard grounded in the Convention on the Rights of Persons with Disabilities (CRPD) would need to require. Under the Chatham House Rule, we are not naming participants or attributing anything that was said.

Why IDA was there

With AI, many of the rules are not written by governments but in requirements, documents, procurement contracts and training data, often inside companies. Whether a benefits system flags a person as a risk, or whether a deaf person can trust the captions on a call, it is settled there long before anyone sees the product. Governments are not yet filling the gap: a forthcoming G3ict report found that only 34 of 193 UN Member States publish enough information to show how they govern AI in public services for persons with disabilities. Article 4.3 of the CRPD requires States to involve us, through our representative organisations, in decisions that affect our lives.

‍

What we heard

One line from the morning summed it up: “No one is ever accidentally accessible.” Every table described the same pattern:

1. Accessibility is still not built in when AI is designed. Persons with disabilities are missing from design teams, testing, and AI standards.

2. Failures are found after launch by the people they harm, because disability data is missing or out of date, and systems are not transparent about where they break.

3. AI systems must be tested for their impact on persons with disabilities before release, not repaired afterwards.

What the group agreed

Participants left with a first draft of five minimum elements for CRPD-aligned AI governance:

1. Impact testing before release

2. Persons with disabilities involved from the outset

3. Accessibility as a condition of procurement

4. Transparency, with a route to appeal

5. Accurate disability data

Tables also set out first ideas for putting these into practice, from a multi-stakeholder working group to international cooperation on standards.

‍

What we are taking back

There is no common reference yet. Engineers, procurement teams and regulators each work from their own idea of what inclusion means, and nobody can be held to it.

“Standards take years to write, and AI changes in months, so the reference has to be practical enough to use now and built to be updated. A reference grounded in the CRPD would give them the same starting point. IDA’s members are well placed to help write it together with industry and standards bodies.” — José María Viera, Executive Director, IDA.

Disability inclusion needs the private sector, but partnership must never become substitution. For companies, that means bringing organisations of persons with disabilities (OPDs) in while requirements are still being drafted and testing for impact before release. For our movement, it means coming up with concrete proposals and saying what would be enough, as well as what is wrong. Read José María's full reflection.

‍

What comes next

The outcomes of Arlington are the starting point for IDA’s work with industry on AI and will feed into preparations for the Global Disability Summit 2028 in Doha. If your organisation builds, buys or regulates AI, which OPDs are you working with? If you are not working with one yet, IDA can help you make that connection.

Our thanks to G3ict and to everyone who came prepared to talk openly about the systems they work with.

‍