For decades, a chip that lets you move a cursor with a thought lived in science fiction. In 2024 it moved into operating rooms. Since then, several companies have placed neural implants in living human volunteers, Neuralink has expanded its study across four countries, and academic teams have published peer-reviewed results letting people with paralysis "speak" through a computer. This is a story about hardware, companies, trials, and the rules being written around them — not about the biology of any one disease, and not about artificial intelligence in general.
The temptation is to read every announcement as a leap toward reading minds. The reality is narrower and, in its own way, more remarkable: small numbers of profoundly paralyzed people are using implanted electrodes to do specific things — move a cursor, type, produce speech — under careful medical supervision. Sorting the measured results from the marketing is the whole task.
What a brain-computer interface actually is
A brain-computer interface (BCI) is a device that records electrical activity from the brain and translates it into a command for a computer or other machine. An implanted BCI usually sits on or in the motor cortex, the strip of brain that plans movement. When someone who is paralyzed attempts or imagines a movement, those neurons still fire; the implant reads the pattern, and software maps it onto an action — a cursor moving left, a letter being selected, a synthesized word.
The medical targets are specific and serious: paralysis from spinal-cord injury or stroke, and neurodegenerative conditions such as ALS (amyotrophic lateral sclerosis) that can leave a fully aware person unable to move or speak. For these patients, "control a cursor" is not a party trick; it can mean the return of independent communication. That framing matters, because it separates the genuine clinical work from the consumer "mind-reading" hype that often rides alongside it.
Implants differ mainly in how invasive they are, and that trade-off runs through the whole field. Penetrating microelectrode arrays push tiny needles into brain tissue and capture the sharpest signals from individual neurons. Surface arrays lay electrodes on top of the cortex without piercing it, trading some signal resolution for lower surgical risk. And at least one approach avoids opening the skull altogether. More signal generally means more capability but more risk; less invasive generally means safer but coarser. No design has "won."
The companies racing to put electrodes in human brains
The most famous name is Neuralink. Its PRIME study (Precise Robotically Implanted Brain-Computer Interface) places the N1 implant in the motor cortex using a surgical robot, and the company reports that participants with quadriplegia have controlled cursors, played games, and typed by intention [source: Neuralink, 2025]. Neuralink says it has moved from its first human implant in early 2024 to more than a dozen participants and has begun enrolling internationally, including UK and UAE arms registered on public trial registries [source: ClinicalTrials.gov, 2025]. These are important to label clearly: the participant counts and performance figures are company statements, not peer-reviewed clinical publications. They may well hold up, but independent, published validation of Neuralink's numbers lags behind its announcements.
Synchron takes the least invasive route. Its Stentrode is threaded through a blood vessel — up the jugular vein and into a large vein sitting against the motor cortex — so it can be placed without open-skull surgery. In the US COMMAND early feasibility study, six patients with severe upper-limb paralysis received the device; over twelve months, all six met the primary safety endpoint, with no device-related serious adverse events causing death or permanent increased disability, and the company reported the device was deployed on target in every case [source: Synchron, 2024]. That is a real, prospective safety result — but note the scale (six people) and that efficacy was described qualitatively rather than proven in a large controlled trial.
Precision Neuroscience is pursuing the surface-array path and reached a regulatory milestone first: in April 2025 the FDA cleared its Layer 7 cortical interface — a film thinner than a hair carrying 1,024 electrodes — to record and stimulate for up to 30 days [source: Precision Neuroscience, 2025]. Clearance of a limited-duration electrode array is not the same as approval of a complete, permanently implanted therapeutic system, but it is a concrete regulatory step. Paradromics placed its Connexus device in a person temporarily during an epilepsy surgery in 2025, then in June 2026 the University of Michigan performed the first implantation for an FDA-approved early feasibility study; the array carries 421 microelectrodes and transmits wirelessly to a transceiver in the chest [source: Michigan Medicine, 2026].
Behind these companies stands a quieter foundation. Blackrock Neurotech makes the Utah Array used in much of the field, and the academic BrainGate consortium — Brown University, Massachusetts General Hospital, Stanford, UC Davis and others — has produced two decades of peer-reviewed BCI research on cursor control, robotic arms, and speech. When you want to know what has actually been measured rather than announced, the academic literature is usually where to look.
The clearest wins are in speech
The strongest evidence so far is not a flashy demo; it is a pair of peer-reviewed speech studies. In 2024, a team at UC Davis, working with a 45-year-old man with ALS, used 256 cortical electrodes to decode his attempted speech with roughly 97% accuracy over a 125,000-word vocabulary, published in the New England Journal of Medicine [source: New England Journal of Medicine, 2024]. In 2025, a follow-on study in Nature went further, synthesizing an audible voice from his neural activity almost instantly — turning raw brain signals into sound within about a hundredth of a second, with closed-loop feedback so he could hear himself [source: Nature, 2025].
These results earn the word "breakthrough" more honestly than most, because they cleared peer review and reported hard numbers. But two caveats keep them grounded. First, each study involved a single participant; high performance in one person is a proof of concept, not a population result. Second, this is decoding attempted speech from someone who wants to speak — it is emphatically not reading private, unspoken thoughts. The distinction between decoding a deliberate communication attempt and "reading a mind" is the difference between the science and the hype.
Reading the numbers carefully
Because the field mixes peer-reviewed science with competitive product announcements, the same three habits that serve any technology story serve here.
Separate company claims from independent validation. A press release reporting that participants type quickly or hit a certain cursor "bit rate" is a company describing its own results; a paper in Nature or NEJM has passed outside scrutiny. Both can be true, but they carry different evidential weight, and right now the peer-reviewed evidence base is dominated by academic and hospital teams, while some of the most-quoted performance figures come straight from companies [source: Neuralink, 2025].
Respect small numbers. Almost every result in this article — six patients here, one participant there — comes from early feasibility studies whose first job is safety, not proof of benefit. Early feasibility is exactly what it sounds like: a first careful look in a handful of people. It is how responsible device development starts, and it is also why sweeping claims about what BCIs "can do" are premature.
Be careful with cause and comparison. When a participant improves at a task over weeks, some of that gain is the device and some is learning, practice, and better software tuning; performance figures are not all attributable to the hardware alone. And a number from one company's setup cannot be directly compared with another's, because the tasks, implants, and patients differ. Impressive demonstrations are real, but they are demonstrations, not head-to-head evidence.
The regulatory path, in plain terms
No fully implanted, general-purpose BCI has FDA marketing approval as of 2026. Every human implant discussed here is happening under an investigational framework, and understanding that framework explains a lot of the news.
The usual sequence runs like this. A device can receive an FDA Breakthrough Device Designation, which speeds up interaction with regulators for serious conditions — Neuralink obtained such designations, including one for a speech application in 2025, and Synchron received one back in 2020 [source: MassDevice, 2025]. Designation is not approval; it is a fast lane, not a finish line. To implant in people, a company needs an Investigational Device Exemption (IDE) to run an early feasibility study — the small, safety-first trials described above. Positive early results lead toward a larger pivotal trial; Synchron has said it is preparing one, the step it must clear before seeking a premarket approval (PMA) to sell a device [source: Synchron, 2024]. Only at the end of that road does a BCI become an approved medical product.
The pathway is also widening beyond paralysis and ALS. In 2026, CorTec's system received a Breakthrough Device Designation for stroke motor rehabilitation, reportedly the first BCI so designated for that use [source: CorTec, 2026] — a reminder that "BCI" is a category of very different devices and indications, not a single product.
The ethics: neural data, reversibility, and long-term safety
The hardest questions are not purely technical. The most novel is neural data: signals recorded directly from the brain are among the most intimate data imaginable, and the law is only beginning to respond. In 2024 Colorado became the first US state to add neural data to its privacy law, and California extended its consumer-privacy act to cover neural data effective January 2025; both generally require opt-in consent to collect and process it, and other states have introduced similar bills [source: ArentFox Schiff, 2024] [source: Arnold & Porter, 2025]. Supporters argue this pre-empts a genuine harm before consumer neurotech scales; skeptics note the definitions are new and untested, and that medical-research data is often handled under separate rules. It is an unsettled debate, not a solved problem.
Then there is reversibility and long-term safety. An implant is not a phone you upgrade; removing or replacing one is surgery, and tissue can react to a foreign body over years. Early studies emphasize safety endpoints precisely because the long-term record is still being written — the multi-year data simply does not exist yet for most of these devices. Less invasive designs, such as Synchron's endovascular approach or surface arrays, are partly a response to that worry, trading some capability for lower surgical and explantation risk.
Finally, the evidence itself is thin by design. Small trials cannot tell us how these systems perform across thousands of diverse patients, how durable the benefits are, or how often they fail. Enthusiasts see an inflection point; cautious clinicians see promising pilots that must still prove themselves at scale. Both readings are defensible, and honest coverage holds them together rather than picking the exciting one.
What to watch
Three things will show whether 2024-2026 was a true turning point or an early chapter. First, watch for peer-reviewed clinical results from the companies, not just press releases — independent publication of implant performance and durability is the signal that matters most. Second, watch the regulatory ladder: whether an early feasibility study graduates to a pivotal trial and, eventually, the first premarket approval for an implanted BCI. Third, watch the rules, from neural-data privacy laws to standards for device safety, reversibility, and informed consent, since those will shape who can build these systems and how.
The measured story is genuinely hopeful and genuinely unfinished. People who could not communicate are, in careful trials, doing so again. That is worth taking seriously — and so is the distance still to travel between a handful of implanted volunteers and a proven, approved medical technology. Watch the peer review, watch the trials, and watch what regulators actually clear, not just what gets announced.