For years, the most useful accessibility features came with a hidden tax. You pointed your camera at a prescription label, and a recording of it travelled to a company server, got processed there, and came back as speech. The captions on a video that shipped without any, the photo read aloud, the menu described out loud: all of it made a round trip through a data centre you could not see. That trip cost you privacy and a working data connection every single time.
In the spring of 2026, both Apple and Google moved a large share of that work onto the device itself. The chip in your hand now does what a server used to do. That is the real story of this announcement cycle, and it matters to people with disabilities more than to almost anyone else.
Here is why. Accessibility tools handle the most personal data a person owns. What a blind user points a camera at. What a Deaf user is watching. What a person with a speech disability is trying to say. Keeping that on the device is not a marketing nicety. It is the difference between a tool you can trust with a doctor’s letter and one you cannot.
No spam. No inspiration porn.
Our best writing for adults with disabilities, weekly and free.
Get the newsletterSo here is what was actually announced, what runs now, and what is still a promise for the autumn.
What Apple announced
Apple previewed its 2026 accessibility features on 19 May, a few days before Global Accessibility Awareness Day and weeks ahead of its developer conference, which ran 8 to 12 June. The headline features are built on Apple Intelligence, the company’s on-device AI system, and process data locally rather than in the cloud.
Four are worth knowing about.
Image Explorer in VoiceOver. VoiceOver is the screen reader built into iPhone, iPad and Mac. It now gives richer descriptions of images: what is in a photograph, a scanned bill, a personal record. Apple says this runs on the device.
Live Recognition in VoiceOver. A user can press the iPhone Action button and ask a question about whatever the camera is pointed at, then ask follow-up questions in their own words. That is a real shift, from a single fixed description to a back-and-forth you steer yourself.
On-device captioning for any video. iPhone, iPad and Mac can generate subtitles for videos that arrived with none, and Apple states the audio is not sent to its servers to do it. Apple puts one limit on it in the footnotes: generated subtitles arrive in English, in the United States and Canada, so this is one of the features that does reach Canadian users at launch.
Vision Pro wheelchair control. The Vision Pro headset can steer a compatible powered wheelchair using eye movement alone, with eye gestures to pause and stop. For people who use a power wheelchair and cannot drive with a joystick, an alternative drive control is not a gadget. It is independent movement. At launch this works with the Tolt and LUCI drive systems, in the United States only.
That last one needs a flag for Canadian readers. Eye-controlled steering is real, but it is United States only at launch and depends on owning a Vision Pro plus a compatible drive system. For most people in Canada it is a preview of where the technology is heading, not something to buy this month.
What Google announced
At Google I/O 2026, the company went so far as to call Android 17 an “intelligence system.” Gemini, Google’s AI model, sat at the centre of the keynote, and Gemini Nano now runs a capable version of it directly on the phone’s processor, no cloud round trip required.
Two threads matter for accessibility. The first is that the same on-device processing behind Live Caption and image description means less of your data leaves the phone, and these features keep working on a weak connection or none at all. The second is quieter and useful: Google extended Live Threat Detection to watch for apps that misuse accessibility permissions. Those are the overlays and screen-reading powers that assistive apps legitimately need and that malicious apps sometimes abuse. A security improvement aimed squarely at the tools disability communities rely on every day.
Google’s reveal was lighter on brand-new, named accessibility features than Apple’s this cycle, and heavier on the underlying plumbing. That is worth saying plainly. The platform-level move to local AI is the story, not a single flashy tool.
The privacy angle, plainly
On-device processing is not automatically private, and it pays to understand why. A feature can run locally and still phone home with usage data. “On-device” describes where the computation happens. It is not a promise about everything else the app does. Read what each company actually commits to, not the word on the slide.
The direction, though, is the right one. For years, the most capable accessibility AI required sending your surroundings, your reading, and your conversations to a server. Moving the heavy work onto the device removes a real barrier: the worry that using the tool means handing over sensitive information. For a person who needs a camera to read every piece of mail, that worry was never paranoia. It was a reasonable read of how the technology worked.
Real now versus later this year
This is where the marketing and the calendar part ways. Apple was clear that its 2026 features are coming later in 2026, with developer betas after the June keynote and public releases following in the autumn. So as of June, most of what was announced is previewed, not shipping. If a feature matters to you, the practical move is to wait for the public release, then test it yourself before you rely on it.
Google’s on-device features roll out through Android 17 and Gemini updates across the year rather than landing all at once. Live Caption and on-device description shipped in earlier forms already; the 2026 work deepens them.
What to do now
Three concrete steps.
First, when these updates reach your device, turn the feature on and test it against a real task you care about. A prescription label. A video without captions. A scanned form. AI descriptions are improving fast, but they still get things wrong, and a confident wrong answer is worse than no answer when the stakes are real.
Second, check the privacy settings that ship alongside each feature. Look for whether processing happens on-device, whether anything is sent off for “improvement,” and whether you can turn that sharing off.
Third, do not pay to upgrade hardware on the strength of a feature that has not shipped. The eye-controlled wheelchair steering is genuinely promising, but it is United States only, headset dependent, and early. Promising is not the same as available. Wait for the Canadian rollout and for independent reviews from users with the same access needs as you.
The bigger picture is encouraging. The most powerful accessibility tools are moving onto the device, where your data stays with you. That is a structural change, not a gimmick. Just keep the calendar honest. A lot of June’s announcements are autumn’s reality.
Sources
- Apple Newsroom, “Apple unveils new accessibility features, and updates with Apple Intelligence” (19 May 2026)
- TechCrunch, “Apple announces Apple Intelligence-powered accessibility feature updates” (19 May 2026)
- Engadget, “Apple’s new accessibility feature lets Vision Pro users control a wheelchair with their eyes”
- Next Reality, “Apple Vision Pro Wheelchair Control Explained: Features and Limits”
- MacRumors, “WWDC 2026: Everything to Expect” (keynote and conference dates)
- Android Central, “Google I/O 2026 Live Blog: Android 17, Android XR glasses, Gemini AI news”
