Researchers at the Weizmann Institute of Science have built a model that rebuilds the picture in front of a person’s eyes from a brain scan (The Times, October 2026). The system, called Brain-IT, reads functional MRI recordings and hands its estimate of the scene to an image generator. Earlier methods needed about forty hours of scanning per volunteer before they produced anything useful. Brain-IT reaches comparable results from one hour (Beliy et al., ICLR 2026).
The headlines call it mind reading. The work reconstructs images a person is actively looking at, and it does not pull dreams or private thoughts out of anyone.
Two details in the method matter more than the headline. The first is that the final picture is partly a guess. The scan supplies a coarse layout and a sense of what the scene contains, and a diffusion model fills in everything the brain signal could not specify. What comes out is the person’s perception passed through the machine’s expectation of what such scenes usually look like.
The second is that the decoder has a partner running the other way. The team trained an encoder that takes an image and predicts the brain activity it should produce, and they grouped the scan data from every volunteer into 128 shared functional clusters so one model could serve them all. A common map of perception across different heads is the groundwork for moving a view from one of them to another.
That move has already been tried at small scale. In 2019 a University of Washington team linked three people through a system called BrainNet (Jiang et al., Scientific Reports, 2019). Two senders wearing electroencephalography (EEG) caps decided whether a falling block in a Tetris-like game should rotate. Their decisions traveled over the internet to a receiver, whose visual cortex was stimulated with magnetic pulses. Five groups played and averaged about 81 percent accuracy. The bandwidth was a single yes or no per round, and it was still three brains working one problem through a channel.
Groups also already live on remote worlds through a channel. When a rover lands on Mars, the operations team at NASA’s Jet Propulsion Laboratory has traditionally spent the first months on Mars time, starting work about forty minutes later each day to keep pace with the planet’s longer sol (Mastcam-Z team, Arizona State University). Ahead of the 2004 twin-rover landings, team members rented apartments with blackout shades and ordered custom watches that ran the extra 39 minutes (Cornell Chronicle, 2003). None of them left Earth. A whole team kept the hours of a place reached only by radio, and their bodies carried the jet lag of a planet they never touched.
In the Mobius Nexus Cycle the Lenses and the Uplink are the two halves of that arrangement. The Lenses sit in the eye and render the network directly into perception. The Uplink is the channel that carries what they render. Contact with the distant world at the center of the series runs through that channel alone, across roughly thirty billion light-years, and no one in the books ever physically travels there. A group that experiences that world together is sharing a rendering.
Brain-IT shows why the word rendering carries weight. A reconstruction is the incoming signal plus a model’s prior. Scale that to a room of people looking at another planet through the same Lenses and the scenery they agree on came partly from the far end and partly from whatever filled the gaps on the way in. Agreement inside the room shows only that everyone received the same picture.
Security practice has a name for half of this problem. NIST SP 800-53 control SC-8, mirrored in ITSP.10.033, asks that information be protected from unauthorized change while it moves. Encryption and checksums can show that the signal arriving at the far end matches the signal that left. They have no way to show whether the model that turned that signal into a picture added something of its own. Once perception itself travels, the decoder becomes part of the channel, and its training becomes something to audit along with the wire.
The Mars teams always knew which clock was which, because two watches kept the planets apart on their wrists. A Lens has one display. A group standing together on a world none of them will reach will have to decide how much of the view to trust, and who trained the part they cannot see.
RECORD RETAINED
SOURCE INTEGRITY UNCONFIRMED