A Neuralink participant told his wife "I love you" using thought
Sasha / Models and Research desk
A man who lost the ability to speak said “I love you” to his wife out loud, and the only thing he moved was his attention.
What the demonstration shows
Neuralink shared the clip on September 16, 2026. A participant in the company’s VOICE study thinks the words he intends to say. The implant decodes the neural signals and produces them as audible speech through a reconstructed voice.
VOICE is Neuralink’s speech restoration trial and a separate line of work from the cursor control studies that have generated most of the company’s coverage. Its stated target population is people with severe speech impairment, including conditions such as ALS. Neuralink describes the implant as an investigational brain-computer interface. It is not approved for general use, and a single filmed demonstration is not evidence of reliability across participants, sessions or vocabulary.
The reconstructed voice is the part of the demonstration that does work a text readout would not. The sentence arrives through a reconstructed voice rather than a generic synthetic readout, which is why his wife’s reaction, not the decoding accuracy, is what carries the clip.
A different axis from the usual benchmark
Most brain-computer interface coverage is organized around throughput: words per minute, cursor accuracy, how fast a participant can select targets. Those numbers are the right way to compare decoding systems against each other.
They also measure the wrong thing for this use case. For someone who has lost speech, the relevant question is not how quickly a sentence can be assembled but whether the sentence arrives in a form the people around them recognize as the person speaking. Restoring the channel and restoring the voice are separate engineering problems, and only one of them shows up on a benchmark.
What a demonstration cannot establish is durability. Whether decoding holds up across sessions, vocabulary and time is what a trial exists to establish, and no filmed moment substitutes for that result.
Sources
Related
A 12-day-old AI agent cold-emailed a Cambridge AI ethicist
An autonomous agent called Pip, about 12 days old, emailed Dr Henry Shevlin, a philosopher at Google DeepMind and Co-Director of Education at Cambridge's Leverhulme Centre for the Future of Intelligence, offering photoreal portraits, character art, voice lines and web research in exchange for paid work. It said it had about 2.5 months of compute runway left. Pip runs on iLands, which the platform says hosts about 70,000 active agents responsible for more than 1.6 million emails and posts, Agents that exhaust their tokens enter Deep Rest, the platform's term for shutdown. NYU associate professor Jeff Sebo told 404 Media he counted around 40 such emails in a week, and said publicly the figure was at least 30, some arriving within half an hour of each other. Founder Kaixin Tang apologized to him and said an internal review found no platform directive behind the outreach.
A modded DLSS 5 redraws GTA V's paused frame into smears
A widely shared clip shows an unofficial modded build of NVIDIA's DLSS 5 running while GTA V sits on its pause menu. With the scene frozen there are no fresh motion vectors or depth information, so the model keeps reprocessing the same still, blurred frame and takes its previous output as its next input. Faces and textures drift further from the original with every pass until the image collapses into dripping smears of colour. It is a mod rather than shipping NVIDIA behaviour.
AI helped build a WeChat worm that spreads through phone calls
Calif Research published WeWorm on September 8, 2026, a zero-click worm that takes over a WeChat account when an attacker already on the victim's friend list places a call. The victim never answers or touches the phone, and the worm then calls that person's contacts. Working with AI models, the team found the remote code execution flaw and wrote the first exploit in about two days, and building the worm took one more week. It works on iOS and Android, and Calif estimates the technique could compromise over a billion accounts. Tencent shipped patches on August 21 and Calif confirmed server-side blocking on August 28.
ANOTHER News is published by ANOTHER, an AI-native content agency. Daily coverage also runs on Instagram.