As of May 2026, implanted brain–computer interfaces (BCIs) are no longer defined only by headline-grabbing first implants; multiple parallel clinical programs now report sustained home use, higher-bandwidth communication, and early “digital bridge” approaches that reconnect motor intent to downstream effectors such as spinal stimulators, while visual prosthesis trials continue to mature with multi‑year safety follow-up and measurable functional gains in controlled tests.[1–3]
Summary
- speech and text output systems for ALS and severe paralysis have moved from proof-of-concept phrases to faster, lower-latency streaming speech and large-vocabulary decoding;
- brain–spine interfaces and adjacent approaches have demonstrated that cortical signals can be linked to spinal stimulation to restore standing and walking in selected cases;
- motor BCIs for everyday device control are being implanted via both intracortical and endovascular routes with year-scale follow-up in some cohorts; and
- visual cortical stimulation systems have reached six-year early feasibility datasets while other cortical-vision teams report multi-participant demonstrations of artificial vision patterns in profoundly blind people.[1, 2, 4–7]
Beyond the cursor
Several independent groups have now shown that speech-related neural activity can be decoded into usable communication outputs in people with ALS and other causes of severe paralysis, using both intracortical microelectrode arrays and subdural or epidural electrocorticography (ECoG).[4, 5, 8]
At UC Davis within the BrainGate2 ecosystem, an ALS participant (SP2) used an intracortical “brain-to-text” neuroprosthesis that, after 30 minutes of training data, achieved a 0.44% word error rate (WER) on evaluation sentences from a 50-word vocabulary in closed-loop mode.[9] When the vocabulary was expanded to over 125,000 words, the same participant achieved a 9.8% WER after collecting 1.9 hours of additional training sentences.[9] In later sessions, average Copy Task decoding in the final three sessions reached 2.66% WER at a self-paced speaking rate of 32.9 words per minute.[9]
Other intracortical speech-to-text work in BrainGate2 has demonstrated faster “conversational-speed” decoding in an ALS participant (T12), reporting 9.1% WER for a 50-word vocabulary and 23.8% WER for a 125,000-word vocabulary at an average pace of 62 words per minute.[10]
ECoG-based speech synthesis has also advanced toward more naturalistic, streaming outputs: in the BRAVO clinical trial, a participant using a 253-channel high-density ECoG array silently attempted sentences drawn from a 1,024-word vocabulary, while the system streamed predicted speech as she began attempting to speak.[4] In online testing, the BRAVO team reported a median decoding speed of 47.5 words per minute and median speech-synthesis latency of 1.12 seconds for the 1,024-word General set, alongside a low false-activation burden in rest data (the system never falsely decoded speech in 16 minutes aggregated across ten sessions).[4]
Finally, Johns Hopkins investigators reported a clinical trial participant with ALS who used self-paced silent speech commands to control smart devices via a chronically implanted ECoG BCI, with a median decoding accuracy of 97.10% across the study period and median online false positive and false negative rates of 0 in detection metrics.[8]
Digital bridges
The most concrete “digital bridge” concept in humans remains the brain–spine interface (BSI) demonstrated by the Lausanne/EPFL team: in a 2023 Nature report, the authors described restoring communication between the brain and spinal cord with a fully implanted system linking cortical signals to epidural electrical stimulation targeting spinal cord regions involved in walking, enabling an individual with chronic tetraplegia to stand and walk naturally in community settings.[2]
A closely related translational pathway is being pursued by ONWARD Medical, which frames its ARC-BCI as pairing a motor-cortex implant with the company’s implanted spinal cord stimulation platform (ARC-IM) to create the ONWARD DigitalBridge, using AI to decode movement intention and translate it into movement.[11] ONWARD reported that its ARC-BCI system received FDA Breakthrough Device Designation in February 2024.[11] In May 2025, ONWARD announced that two additional spinal cord injury procedures brought its total number of successful ARC-BCI implants to five (performed at CHUV in Lausanne, Switzerland).[11] By January 2026, ONWARD reported two additional spinal cord injury implants, bringing the total number of human ARC-BCI implants to seven, again at CHUV under the direction of neurosurgeon Jocelyne Bloch.[12]
A different, non-implant “bridge-like” approach in spinal cord injury is also emerging in controlled studies: a 2026 randomized pilot trial (ChiCTR2300074503) in 21 people with spinal cord injury compared BCI-controlled exoskeleton training versus exoskeleton-only training, reporting significant within-group gains in walking speed (10MWT, ) and endurance (6MWT, ) in the BCI+exoskeleton group, though between-group differences were not significant.[13]
Bionic limbs and motor neuroprostheses
For everyday computer and device control, Neuralink and Synchron illustrate two divergent surgical strategies—intracortical threads inserted by a robot versus an endovascular “stent-electrode” array placed via catheter—while academic consortia such as BrainGate continue to demonstrate multi-modal control that blends communication and cursor/robotic outputs in the same participant(s).[14–16]
Neuralink told Reuters in January 2026 that it had 21 total participants enrolled in trials worldwide, up from 12 reported in September 2025, and described maintaining a record of zero serious device-related adverse events while working with regulators and hospital sites.[1] Reuters also reported that the first patient used the implant to play video games, browse the internet, post on social media, and move a cursor on a laptop.[1] Neuralink’s own updates describe its Telepathy concept as translating neural activity from hand/arm motor regions into digital commands, and report that one participant (“Nick”) achieved over 10 bits per second within his first week of BCI use and later used a robotic arm to complete basic tasks such as feeding himself and scratching an itch.[17]
Synchron’s Stentrode platform, by contrast, is implanted in a blood vessel on the surface of the motor cortex via the jugular vein and is designed to detect and wirelessly transmit motor intent for hands-free point-and-click control of personal devices.[18] In September 2024, Synchron announced positive 12‑month COMMAND study results in six patients, reporting that all six met the primary endpoint of no device-related serious adverse events resulting in death or permanent increased disability and that there were no serious adverse events related to the brain or vasculature during the 12‑month period.[18] Synchron also reported 100% accurate deployment achieving target motor cortex coverage, with a median deployment time of 20 minutes.[18] By late 2025, Synchron stated that Stentrode BCIs had been placed in 10 patients with paralysis across clinical trials in the U.S. and Australia.[19]
The table below summarizes a few clinically relevant contrasts that recur across these programs.
Restoring sight
In visual neuroprosthetics, the most mature published “program-scale” dataset in this source set is Cortigent’s Orion Visual Cortical Prosthesis System, which has reported multi-year early feasibility follow-up and functional test differences with the system on versus off.[6] Cortigent’s own 6‑year early feasibility summary states that six subjects were implanted between January 2018 and January 2019 and that the study concluded in March 2025.[3] Across that study, Cortigent reported that all devices remained functional throughout follow-up with loss of functionality on fewer than 4% of electrodes, and that one serious adverse event (a seizure) occurred early, with no further seizures or serious adverse events after stimulation patterns were adjusted.[3] In describing the system’s mechanism, Cortigent says the Orion system uses a wirelessly powered implantable pulse generator connected to an array of 60 micro-electrodes on the visual cortex, and that camera input is converted into wireless commands that elicit phosphenes (spots of light).[3]
Independent cortical-vision work also continues: the University of Utah’s Moran Eye Center reported in 2023 that an experimental prosthesis “hardwired into the visual regions of the brain” had been used to safely provide a form of artificial vision to three individuals with blindness, and reported that Eduardo Fernández described similar results in two additional study participants at a symposium.[7] Separately, a clinical trial case report of intracortical microstimulation described implantation of a 100-electrode Utah Electrode Array near the V1/V2 border in a profoundly blind participant, after which the participant regained perception of light and motion and could read large characters and words.[22]
Neuralink’s “Blindsight” is still at the regulatory-milestone stage in the evidence provided here: sources report that the company’s experimental Blindsight implant received FDA Breakthrough Device status in September 2024 and is intended to restore vision by directly stimulating the visual cortex.[23]
Remaining challenges
Despite striking performance demonstrations, much of the strongest evidence base remains small-N or single-participant, as underscored directly in implanted speech-BCI reporting that notes the key limitation of single-participant studies and the need to replicate results in more participants.[24] Long-term reliability is also not uniform across interfaces: one chronic ECoG gesture-decoding report in an ALS participant found offline classification accuracy declined from 49.3% to 28.0% across two periods separated by roughly five months, alongside reported reductions in high-gamma band power modulation and increased false positive frequency.[25] At the same time, multiple programs emphasize structured safety oversight (e.g., FDA IDE frameworks), which is central to scaling implant counts beyond early feasibility cohorts.[8, 18] Finally, ethical and transparency pressures remain salient in high-profile commercial efforts; one analysis argued that not registering Neuralink’s first clinical trial in ClinicalTrials.gov “seemed to violate” fundamental ethical guidelines, even though it reports the record was later submitted in May 2024.[23]
