Here's another single-article newsletter. The topic matters, and there was no way I was going to do it justice in a few hundred words.

I first heard about this while listening to The Accessibility Breakdown podcast. The part that got to me was about a post by Bianca Prins about Meta donating 15,000 smart glasses in Ireland. I was driving into town to run some errands, and by the time I was heading home, my brain had started pulling at a thread. I dictated the bones of this piece on the drive back.

The next day I started filling in the gaps, checking some of the things I'd been thinking about and following a few connections I hadn't expected when I started.

And, well... here it is. Enjoy the read!

Let's dive in.

In this newsletter

1. How do you govern a disability dataset you're told doesn't exist?
2. Wrapping up

How do you govern a disability dataset you're told doesn't exist?

There's an old poker expression: look around the table. If you can't spot the pigeon, it's probably you. The pigeon is the player everyone else expects to profit from.

On August 12, Meta announced that it was giving 15,000 pairs of Ray-Ban Meta smart glasses to Vision Ireland, enough for every blind and visually impaired adult the organization supports. Meta is also funding training, while Vision Ireland will manage the rollout and support recipients.

It's an extraordinary gift. Meta has effectively bought 15,000 people into the game, paid for the equipment and covered the lessons. And the players can absolutely win.

For a blind person, smart glasses can read text, identify objects, describe what's around them and connect with services such as Be My Eyes. Some of those things aren't new. A blind person can already point a phone camera at something and use Be My Eyes or another service to get information about it.

Glasses change the interaction. The camera is already positioned roughly where the person is looking. Asking for information can become part of moving through the world instead of a separate act of taking out and positioning a phone. Be My Eyes describes the glasses integration as a hands-free way to connect with a sighted volunteer. For some people, those glasses are going to be genuinely transformative.

I understand why Meta would want to learn from how people use them, too. If you're building assistive AI, knowing where the technology succeeds and fails is incredibly useful. You can't improve a product without understanding how people use it.

There's another benefit to Meta that's harder to ignore. Smart glasses raise uncomfortable questions about surveillance, recording and privacy. Accessibility gives the technology a pretty compelling answer to the question of why anyone would want a camera on their face in the first place. It's harder to dismiss the glasses as a creepy novelty when the same device lets a blind person read a menu or find their way around somewhere unfamiliar.

I don't know whether this donation is generosity, public relations, or some mixture of the two. I can't know Meta's motives. I can look at what Meta gets from the arrangement. In a staking deal, someone else puts up your buy-in, you play, and the backer gets a share of what you win. Meta gets something too.

I should also be clear about where I'm standing in this discussion. I'm a disabled disability rights activist. I'm not blind, I'm not in Ireland, and I'm not speaking for blind people in Ireland. They are perfectly capable of speaking for themselves, and they may reach very different conclusions about this program than I do.

Nothing About Us Without Us operates at more than one level. Blind people in Ireland should have decision-making authority over a program involving their lives and their data. Disabled people more broadly also have a stake in how disability itself gets defined and measured. A method developed to infer dependence from blind people's product use doesn't necessarily stay with blind people. It can become a template for measuring dependence in other disabled populations, and that's the part I'm speaking about.

This is bigger than privacy

There are obvious privacy questions about putting a camera and microphone on someone's face. What's recorded? What's retained? Where is it stored? Who sees it? What happens when someone standing nearby hasn't consented to being recorded?

I'm interested in another question as well: what can be inferred. Imagine asking your glasses to read a label in a grocery store, then identify the product beside it. Later you ask them to describe a bus stop. You use them to find a doorway. At home, you ask them to identify something on the kitchen counter.

None of those interactions necessarily says much by itself. Put thousands of them together and patterns begin to emerge. How often is the technology used? What is someone asking it to do? Where and under what circumstances?

If someone asks their glasses to identify a product, that does not tell us they couldn't identify it another way. Maybe the glasses were faster. Maybe another technique was inconvenient. Maybe they simply preferred using the glasses. "Asked for assistance" and "was incapable of doing this unaided" are not the same fact.

There are four different things here that I don't want to blur together: what the glasses observe, what Meta actually retains, what could be inferred from whatever is retained, and what someone might eventually decide to do with those inferences. I'm interested in the possibility of moving from one stage to the next, not claiming that movement has already happened.

Governments already have enormous datasets about disabled people. So do health systems, insurers and researchers. Meta has not invented disability data.

What seems different is the potential granularity. Medical records tell you about encounters with a health system. Benefits records tell you about eligibility. Surveys tell you how people answered questions someone thought to ask. A device worn on someone's face potentially encounters disability as it is lived, minute by minute.

What does Meta actually collect?

Meta has privacy policies describing what its glasses collect and what happens when people use Meta AI. The problem is that those policies can change.

In April 2025, Meta changed the privacy settings for Ray-Ban Meta glasses in the United States. AI features, including Meta AI with camera, were enabled by default, and US users who wanted to continue using voice features could no longer opt out of storing voice recordings. Those recordings can be retained for up to a year and used to improve Meta's products, including its AI models. European users retained greater control.

US users received one set of choices. European users retained another. Enforceable legal constraints made a difference.

Meta's history gives me reason to pay attention to the gap between what a product does today, what the technology is capable of doing, and what users reasonably understand is happening. The Irish Data Protection Commission fined Meta €1.2 billion in 2023 over transatlantic data transfers. More recently, questions have involved the glasses themselves.

In Bartone v. Meta Platforms, filed in California in March 2026, plaintiffs alleged that footage from Meta's AI glasses was being reviewed by human contractors, including intimate footage of people undressing, using bathrooms and having sex. The allegations have not been adjudicated. Meta has said faces in footage sent for review are blurred, while workers interviewed by journalists said that blurring did not always work reliably. The reporting also prompted the UK Information Commissioner's Office to investigate.

Then there's face recognition. In February 2026, reporting based on internal Meta documents said the company was considering introducing facial recognition to the glasses. One internal memo reportedly discussed launching it during a "dynamic political environment," when civil-society groups expected to criticize the feature would have their resources focused elsewhere. In June, WIRED found that the machinery for a feature called NameTag had already been built into the Meta AI app, although Meta maintained that no final decision had been made to release it. Meta removed the code after WIRED's reporting.

Face recognition brings us straight back to accessibility. The reported Meta plan included the possibility of first releasing NameTag at a conference for visually impaired people. Helping a blind person identify someone in front of them is an obvious potential benefit. It's also a powerful justification for introducing a capability that raises serious privacy questions for everyone whose face comes within view of the camera.

Bianca Prins has made a related argument from a European policy perspective. She argues that giving 15,000 disabled people access to technology that can increase their independence changes the political calculation if European governments later consider restricting that technology over privacy or surveillance concerns. A restriction could then be framed as taking away an accessibility tool disabled people have come to rely on. In her view, that creates a collision between disabled people's independence and other people's privacy rights, potentially giving Meta leverage in its disputes with European regulators. Her word for what that could make disabled people is "pawns."

Her concern and mine aren't quite the same. She's asking what political leverage Meta gains when disabled people come to depend on the glasses. I'm asking what might eventually be learned about disabled people's lives through their use of them, and who gets to decide what that information means. Both questions come back to how much power disabled people actually have in the arrangement.

This is why I'm uncomfortable treating what the glasses do today as the boundary of what we need to govern. These examples give me no evidence that Meta is secretly collecting disability data from people using the glasses. They do give me a reason to want something stronger than corporate assurances.

People receiving these glasses in Ireland can make decisions about the product that exists in 2026. They can't meaningfully consent today to uses that haven't been invented yet, and I don't think Meta necessarily knows what all those uses will be either.

When does ordinary data become disability data?

Meta could quite truthfully say that it doesn't collect data measuring dependence among blind people. There might be no database column marked DEPENDENCE. There might be no research project examining dependence. Nobody at Meta may even have considered the question.

But perhaps there are records showing that someone asked their glasses for environmental descriptions 18 times, text recognition 12 times and object identification six times. Meta might call that product telemetry, AI interaction data, engagement data or information used to improve a model.

A product analyst sees 36 interactions. Someone interested in functional limitation might see 36 instances of assistance. The blind person might say, "I went grocery shopping."

There's another assumption hiding here. Because these glasses are being given to blind and visually impaired people, it would be easy to interpret every request through blindness. But we don't necessarily know why someone made a particular request.

If someone asks the glasses to identify a person, perhaps they couldn't see the person's face. Maybe they have difficulty recognizing faces, a memory impairment, or simply couldn't remember who the person was. Blind people can have other disabilities, and many people acquire vision loss as they age alongside other impairments.

Knowing that a user is blind doesn't tell us why they used the technology. Frequency of requests, types of requests and repeated tasks can all become proxies. Combine enough of them and someone can start giving the resulting patterns names: independence, functional limitation, social participation, perhaps even quality of life.

When I use the term "disability data" here, I'm using it more broadly than a particular legal category under GDPR. I mean information being used to draw conclusions about disability or disabled people's lives. Who gets to decide when ordinary product data becomes disability data?

The same 36 interactions

Go back to the person whose glasses recorded 36 interactions: 18 requests for environmental descriptions, 12 for text recognition and six for object identification.

A product analyst might see 36 interactions. Someone interested in functional limitation might describe them as 36 instances where the person needed assistance. But now put those interactions back into the context of a person's life.

Maybe she went into town. She used the glasses to find the bus stop, read signs, identify products in a shop and navigate somewhere unfamiliar. When she got home, perhaps she used them to read her mail.

One analysis could describe those 36 interactions as evidence that she needed help repeatedly. Another could say she went about her day and did 36 things without having to ask another person for help.

Same observations, completely different story about her life. Neither interpretation has to be entirely wrong, which is part of the problem. Thirty-six interactions can't tell us what dependence means in her life. Someone still has to choose what conclusions to draw from them.

Nothing about us without us

Disabled activists have been saying "Nothing About Us Without Us" for decades. We usually mean that disabled people must be involved in decisions affecting disabled people. I think data adds something to that principle. We also need authority over what our lives are said to mean.

Someone has to decide what independence and dependence mean, what counts as meaningful activity or participation, and what a good quality of life looks like. More importantly, someone decides which questions are worth asking in the first place.

AI complicates this. AI systems are trained on enormous amounts of material created by human societies. Ableism is deeply embedded in those societies. But we don't even need an AI to spontaneously develop an ableist understanding of disability.

Someone could ask: "Identify behavioural indicators associated with loss of independence."

The assumption is already inside the question. Someone has decided that particular behaviours constitute evidence about independence, before the model analyzes a single record.

Disabled people need more than a seat at the table when someone announces a disability research project. We need authority over the questions being asked, which becomes difficult when nobody acknowledges that the disability dataset exists.

The vocabulary already exists

I should explain why my mind goes where it does with all this.

I've spent nearly 30 years opposing the legalization and expansion of assisted suicide and euthanasia. That experience has made me particularly attentive to the language used to describe disabled people's dependence, suffering and quality of life.

Health Canada's Sixth Annual Report on Medical Assistance in Dying in Canada, covering the 2024 calendar year, is worth looking at closely.

Canadian MAID reporting divides cases into two tracks. Track 1 applies when a person’s natural death is considered reasonably foreseeable. Track 2 applies when it is not. The distinction matters here because some of the differences in the reported sources of suffering are particularly pronounced among Track 2 recipients.

Among people receiving MAID, loss of the ability to engage in meaningful activities was reported in 95.1% of Track 1 cases and 97.5% of Track 2 cases, while loss of the ability to perform activities of daily living was reported in 85.4% and 85.1% respectively.

Isolation or loneliness was reported in 44.7% of Track 2 cases, where natural death was not reasonably foreseeable, compared with 21.9% of Track 1 cases. Perceived burden on family, friends or caregivers was reported in 50.3% of Track 2 cases and 48.4% of Track 1 cases. Loss of dignity was reported in 73.9% of Track 2 cases and 63.5% of Track 1 cases.

Inadequate pain control, or concern about it, was reported in 59.8% of Track 2 cases and 55.9% of Track 1 cases. These are reported sources of suffering among people who received MAID, rather than rankings of why people chose it. It's notable that loss of meaningful activity, difficulty with daily living and loss of dignity are reported more frequently than inadequate pain control or concern about controlling pain.

Even that last statistic contains an interpretive choice. Health Canada reports "inadequate pain control, or concern about it" as a single category. Inadequate pain control and concern about controlling pain aren't the same experience, yet they're combined in the reporting. Even apparently straightforward statistics contain decisions about what to count, what to combine and what to call the result.

Disabled people aren't incidental to these numbers either. Of the 16,104 MAID recipients who answered Health Canada's question about disability, 5,295, or 32.9%, identified as having a disability. Among Track 2 recipients who answered the question, that rose to 61.5%.

Concern about the structural conditions surrounding MAID and disability isn't limited to people like me who have spent decades opposing assisted suicide and euthanasia. In its 2025 concluding observations on Canada, the UN Committee on the Rights of Persons with Disabilities called for Track 2 to be repealed. It also called on Canada to address systemic failures involving poverty, access to health care, accessible housing, homelessness, community-based mental health support, care at home, personal assistance and employment support.

The numbers also need some context. They don't mean people are receiving MAID simply because they're lonely or feel like a burden. Health Canada specifically examined this question in the sixth annual report. Isolation or loneliness was not reported as the sole source of suffering in any Canadian MAID case in 2024; people reporting it averaged 7.5 sources of suffering under Track 1 and 7.1 under Track 2.

Health Canada's reporting describes suffering using a vocabulary that includes meaningful activity, daily living, independence, dignity, isolation and burden. Sufficiently rich observations from smart glasses could also be interpreted along those dimensions.

This doesn't establish a connection between Meta and MAID. It establishes why this kind of data could have political value.

Time for the tinfoil hat

So here's where I put on my tinfoil hat. What happens when anecdote becomes data at population scale?

Assisted-dying advocacy has long made powerful use of individual stories about autonomy, dependence and loss of ability. Stories are powerful, but extremely granular observations about disabled people's everyday lives could turn those stories into something that looks much more objective: data about dependence, independence, dignity, isolation, burden and quality of life.

The interpretive vocabulary already exists while we're building infrastructure capable of generating observations along those same dimensions. I don't know of a pathway by which assisted-dying organizations would gain access to Meta's data. What worries me is what becomes possible once observations like these exist.

Datasets outlive the intentions that created them. Companies change. Ownership changes. Safeguards disappear. Research partnerships form. Analytical techniques improve. Questions nobody thought to ask in 2026 may be easy to answer in 2036.

Meta doesn't have to be collecting the information for that purpose, and nobody there needs to have considered that purpose at all. The information only has to exist while the people generating it have very little control over what questions someone might ask of it tomorrow.

The glasses may prove the opposite

Go back to the Canadian vocabulary: meaningful activity, activities of daily living, independence, isolation. The glasses touch every one of those things, but in the direction of greater independence and participation.

Disabled people have been making this argument forever. Change the environment and provide the accommodation or support, and the things we can do change dramatically.

There's also the person interpreting the observations. They don't arrive without assumptions of their own. They have ideas about what independence looks like and what dependence looks like. And, like the rest of us, they've spent a lifetime in a society full of assumptions about disabled people.

The same device that could be used to construct a story about why disabled lives are unbearable could provide evidence that they need not be. It depends on which story someone decides to tell.

Dependence on whom?

There's another kind of dependence here that I hadn't really thought about until a blind friend pointed it out to me. What happens after someone starts relying on these glasses?

Maybe they now use them to read their mail, shop, get around or identify things every day. Then Meta changes the privacy terms, starts charging for a feature that used to be free, or changes the product in some other way.

"Stop using it" is technically an option. But it isn't much of a choice if you've built part of your independence around the thing.

Blind people have been here before with proprietary assistive technology. JAWS, by Freedom Scientific, one of the most widely used screen readers, is a good example. People built their working lives around it, while the ways they could pay for it and keep it current changed over time. Freedom Scientific introduced annual subscriptions alongside perpetual licences, and more recently changed its maintenance rules so that someone with an outdated perpetual licence can no longer simply buy an upgrade. They have to move to a subscription or buy a new perpetual licence with a maintenance agreement. By then, a user may have years of experience, workplace configurations and habits built around JAWS. Walking away because the terms have changed isn't necessarily a simple choice.

The glasses can reduce dependence on other people while creating a different kind of dependence on Meta. The more useful they become, the more that matters.

Decision-making authority over something that doesn't exist

Saying disabled people need "input" isn't enough. Neither is consultation or putting one blind person on an advisory committee after the research question has already been written.

For this program, blind people in Ireland should have meaningful decision-making authority over disability-related research and inference generated from their use of the technology. But how do you exercise authority over a dataset you're told doesn't exist?

Governance therefore can't apply only to information Meta already recognizes as disability data. It also has to cover what can be inferred from other information.

One way to do that would be an independent data trust governed primarily by blind people participating in the program. It could have actual veto power over research questions, the creation of derived disability measures, the linking of datasets and secondary uses intended to draw conclusions about disability. Its approval would be required, rather than simply sought as advice.

There are other possible models. The important part is who gets to decide, and that authority has to continue as the technology changes. Meta cannot tell recipients every future inference that might become possible because Meta itself probably doesn't know. Recipients also can't know how the product, its pricing or its privacy practices might change after they've incorporated it into their lives. Consent obtained once at the beginning can't cover all of that.

So show us the agreement

There are questions here that don't require any speculation. Publish the data-governance terms of the agreement between Meta and Vision Ireland, along with an accessible plain-language explanation of what those terms mean for the people receiving the glasses.

Tell recipients who controls which data, what is retained and for how long, and the lawful basis for processing it. Explain whether information generated by these users can be used for model training and whether human reviewers can see it. Provide meaningful ways to refuse secondary uses without making the accessibility features less useful.

Vision Ireland is also administering applications, assessment and eligibility for the program. Its program site identifies the glasses as being provided to blind and visually impaired adults the charity supports. The governance of the information created through the rollout should be public too: what Vision Ireland retains, who can access it, whether any of it can be connected to a recipient's Meta account or device data, and what happens to it when the rollout is over.

Put the kind of independent, blind-led governance I described above into the agreement itself, with a say in what happens as the technology, the terms and the things that can be inferred from the data change.

Nothing About Us Without Us shouldn't start only after someone else, very often a non-disabled person or institution, has already decided that what they're doing is about us.

Back to the table

Meta gets something from this arrangement. Thousands of people will use its technology in circumstances that test capabilities most users rarely need. Its models can improve. Its products can improve.

Blind people in Ireland can also gain genuine independence from the glasses. Those things can all be true at once.

I don't object to Meta benefiting from building a better product. I object to disabled people having no meaningful say over what else might eventually be learned about us in the process.

The most valuable thing Meta gets from this arrangement may not be money. It may be knowledge about barriers, assistance, independence and how blind people navigate environments designed around sight.

Maybe my tinfoil hat is showing. I hope that's all this turns out to be.

But history hasn't given disabled people much reason to assume that information about our lives will always be interpreted in our interests. If we're going to generate that knowledge, we should have authority over what it is allowed to become.

Because the pigeon isn't necessarily the player who loses.

It may simply be the player who never got to see the agreement.

Wrapping up

That's it for now! I hope you enjoyed the newsletter. I'd love to get feedback - What was good? What could be improved? What topic would you like me to talk about? I'm not making any promise, but if a topic you suggest catches my fancy, I'll share my opinion on it. Just hit reply to this email, or send an email at [email protected]. I read every response. And a reminder that my content is Human Generated Content #HumanGeneratedContent