The takeaway: After a paralysed patient in China this week became the first person to receive a commercially approved brain-computer interface, the phrase "brain chip" is everywhere — and it's carrying a lot of science-fiction baggage it doesn't deserve. A brain-computer interface, or BCI, is a device that reads electrical activity from the brain and turns it into a command a computer can act on. The useful ones today do one fairly narrow thing well: they let a paralysed person move a cursor, a robotic arm, or a spelling interface by intending to move. They do not read your thoughts, and they can't upload anything into your head. Here's what's actually going on, and why the differences between devices matter more than the headlines suggest.

What a BCI actually reads

When you intend to move your hand, a specific patch of your brain — the motor cortex — fires in a characteristic pattern, whether or not the hand can respond. A BCI eavesdrops on that pattern. Sensors pick up the electrical signal, software learns to map "this pattern means move right," and that mapping drives a cursor or a prosthetic. Crucially, it's reading motor intent — the plan to move — not language, not memory, and not, in any reliable way, the private monologue in your head. The popular image of a device that "reads your mind" is wrong in a specific way: current BCIs read the part of the brain that controls movement, because that signal is legible and consistent. Abstract thought is neither.

The real dividing line: how close you get to the neurons

Every BCI faces the same trade-off — the closer the sensor gets to the neurons, the cleaner the signal, and the higher the medical risk. That trade-off is the entire story, and it's where devices genuinely differ:

• Non-invasive (on the scalp). EEG caps sit on the skin, no surgery. Safe and cheap, but they read the brain's activity through skull and tissue, so the signal is blurry — fine for coarse control, not for fluent, high-bandwidth use.
• Epidural / on the membrane (surgery, but no penetration). The device sits on the dura mater — the tough membrane covering the brain — without entering brain tissue. This is where China's newly-cleared NEO device sits: eight sensors on the membrane, reading the motor cortex from just outside it. You get a better signal than a scalp cap with much lower risk than going into the brain — no penetration means far less chance of haemorrhage or long-term scarring.
• Penetrating (into the cortex). Fine electrodes go into the brain tissue itself. This is Neuralink's approach. It offers the richest signal — you're listening to individual neurons — but it carries the highest risk of bleeding, scarring, and the signal degrading over months as tissue reacts to the implant.

There's no single "best" here. NEO's whole regulatory case is that staying on the membrane is safe enough to approve for commercial use; Neuralink is betting the richer in-tissue signal is worth the harder road. That's the distinction the "brain chip" headlines flatten.

What they can do today — and what they can't

The genuine, demonstrated wins are real and moving: people with paralysis controlling cursors and robotic limbs, and — increasingly — decoding attempted speech to give a voice back to people who've lost one. These aren't parlour tricks; they're restoring function to people who had lost it, and the field is advancing quickly.

The limits are just as real. A BCI trained on your motor cortex cannot read your emotions, your plans, or your memories. And it isn't writing thoughts into your mind. Stimulation can write signals in for specific jobs — restoring a sense of touch to a prosthetic hand, or deep-brain stimulation for Parkinson's — but that is a world away from inserting a thought or a memory, which nobody is doing in any meaningful sense. And every one of these devices requires a real medical procedure with real risk, plus painstaking, individual calibration. "Implant and it works" is not where the technology is.

Why this week mattered

The reason this week's NEO implant was news isn't that it did something Neuralink hasn't — invasive BCIs have been implanted in trials for years. It's that the device had already cleared a national medical regulator for commercial use back in March, and this was the first person to receive it under that approval — moving one of these devices from "experimental trial" toward "approved medical product." That's a regulatory milestone, not a scientific leap, and it's worth watching precisely for that reason: the science has been maturing for a while; the question now is who approves these for real patients, on what evidence, and how fast.

So when you see "brain chip" in a headline, the useful questions are simple ones. Where does the sensor sit — scalp, membrane, or in the tissue? What is it actually reading — movement, or the fantasy of mind-reading? And is this a lab result, a trial, or an approved device? Get those three straight and most of the hype falls away, leaving something that's genuinely remarkable on its own terms.

For this week's news peg, see Saturday's Daily Update on China's first commercial BCI implant.