DeepMind's SL2T brings sign language translation to the Pixel 11 keyboard
Sign language AI just moved into the phone keyboard.
What shipped and where
Google DeepMind released SL2T, a sign-language-to-text model now shipping in Gboard and Live Transcribe on the Pixel 11. It starts with American Sign Language to English, so Deaf and hard of hearing users can sign directly into the keyboard instead of typing.
DeepMind says the model is state of the art on academic benchmarks, and, more usefully, that it is optimized for how people actually sign at a phone: one-handed, while holding the device. That distinction matters because benchmark ASL and real-world phone signing are different problems, and a model tuned only for the former would fail exactly where users need it.
The privacy architecture is notable. Body poses are tracked on-device, and Google’s servers translate those poses into text. Raw video never reaches Google’s infrastructure, only the pose data does.
Built with the community it serves
SL2T was developed with Deaf Googlers and Google’s AI Sign Language Advisory Committee, and DeepMind says more sign languages and applications are coming.
That development process is worth flagging because accessibility features have a habit of being announced in research blogs and then quietly never shipping. This one arrives inside two apps people already use, Gboard and Live Transcribe, on a phone that is actually for sale. Shipping in a default keyboard, rather than in a standalone accessibility app that users have to find and install, is the difference between a demo and infrastructure.
The honest limits
ASL to English is one language pair out of hundreds of sign languages worldwide, and the hard part of extending coverage is that sign languages are not dialects of each other. Each is a distinct language with its own grammar, so every new pair is closer to a new model than a fine-tune.
DeepMind’s stated plan to add more languages is the right ambition, but for now the launch serves ASL users on one device family. Whether SL2T becomes a platform capability or stays a Pixel feature is the thing to watch, because keyboards are where this either scales or stalls.
Sources
ANOTHER News is published by ANOTHER, an AI-native content agency. Daily coverage also runs on Instagram.
Related
An AI newsroom with one human on staff scooped the reporters in the room
RuntimeWire, a tech news site run by AI agents whose founder Ryan Merket is the only human on staff, published details of an OpenAI security disclosure at Black Hat before journalists physically attending the conference filed their stories. Merket spotted a photo from OpenAI president Greg Brockman on X containing a QR code to the talk's live caption stream and fed the transcript to his agents while the session was still running. The site launched in May 2026 and had published 1,627 articles by August 5 out of 71,796 candidate stories evaluated, a 2.3 percent publish rate, at a cost Wired puts at about $100 a day. Reviewers describe the output as flat play-by-play lacking human insight.
Gemini passes 1 billion monthly users, and 63 percent of them use voice
Sundar Pichai announced in a blog post on August 11, 2026 that Google Gemini crossed 1 billion monthly users, making it the fastest-growing product in Google's history and the 14th Google service to pass a billion. 63 percent of users talk to Gemini instead of typing, iOS has more than 100 million active users, over 150 million images are generated a day, and one in five Gemini Live sessions uses live camera or screen sharing. Three weeks earlier, Alphabet's Q2 earnings had put the figure at 950 million. Google did not break out how many users pay.
Three Zoom flaws let a meeting participant take over another attendee's device
Researcher Idan Levcovich of Israeli offensive-security firm A Security disclosed three Zoom vulnerabilities on August 11, 2026, tracked as CVE-2026-53413, CVE-2026-53414 and CVE-2026-53415, that let any meeting participant take over another attendee's device via malformed drawing objects sent through screen-share annotation, with no click or download on the victim's side. Zoom rates two of the flaws 8.3 while A Security rates all three 9.0; patches shipped in June and July for Zoom Workplace, the Workplace VDI client, Zoom Rooms and the Meeting SDK, and no exploitation has been reported. A Security says it built a working exploit using publicly available AI models, fewer than 20 prompts and under 24 hours, though its automated pass over 3,762 functions missed the vulnerable code and a human found it.