Brain-Computer Interfaces in 2026: What Patients Are Actually Doing

On February 20, 2024, Noland Arbaugh — a 29-year-old rendered quadriplegic by a diving accident — live-streamed himself moving a cursor across a MacBook screen using only his thoughts. Less than a month after receiving Neuralink's first human implant on January 29, 2024, he played chess on Lichess, browsed Reddit, and navigated maps without touching a keyboard or trackpad. It was a genuine milestone. In terms of raw information throughput, it was also roughly equivalent to a very slow USB 1.0 connection.

That gap between headline capability and engineering reality defines where brain-computer interfaces stand heading into mid-2026. Three companies are now running active human trials: Neuralink with its penetrating-electrode PRIME study, Synchron with its endovascular Stentrode, and Precision Neuroscience with its surface-mounted Layer 7 Cortical Interface. Each has produced patients doing things unimaginable a decade ago. Each has also revealed hard constraints that press releases tend to omit.

Neuralink's PRIME Study: Progress and a Significant Problem

Arbaugh received his N1 implant on January 29, 2024, becoming the first human subject in Neuralink's PRIME (Precise Robotically Implanted Brain-Computer Interface) study. The N1 chip carries 1,024 electrodes distributed across 64 flexible polymer threads, each thinner than a human hair, inserted by the R1 surgical robot into layer 5 of the primary motor cortex — where motor intent signals are densest. By February he demonstrated reliable cursor control on macOS. By March he had logged multi-hour sessions of online chess. In April he posted on X using decoded neural intent alone, becoming the first person to publish to social media directly via a brain implant. He subsequently demonstrated navigating an Apple Vision Pro, controlling spatial interface elements the headset normally routes through eye tracking and hand gestures.

In May 2024, Neuralink disclosed a material problem: approximately 85% of the 1,024 electrode threads had retracted from cortical tissue in the weeks following surgery, reducing effective electrode count to roughly 150 functional channels. The root cause is micromotion — the brain shifts slightly inside the skull with every heartbeat and head movement, while electrode threads are anchored to the rigid titanium housing fixed to the skull. Sustained differential movement pulls threads free. This failure mode is well-documented in the academic BCI literature; Barrese et al. described chronic microelectrode array degradation in the Journal of Neural Engineering (2013), and comparable retraction effects have been observed in Utah Array implants in both primate and human subjects. Neuralink's flexible polymer threads were specifically engineered to reduce this risk; the May 2024 data showed the problem had not been eliminated. Neuralink engineers retrained their decoding algorithms to extract more signal from the surviving electrodes, and Arbaugh subsequently recovered and reportedly exceeded his pre-retraction functional performance metrics.

Subsequent PRIME Patients

The FDA Investigational Device Exemption (IDE) authorized Neuralink to implant up to 11 patients in the first PRIME cohort. By mid-2025, the company had confirmed multiple additional implants and characterized their recovery trajectories as faster than Arbaugh's, attributing the improvement to iterative protocol and thread-anchoring refinements informed by Patient 1. Per-patient thread retention rates and standardized functional benchmark data have not been released in peer-reviewed form.

What Patients Are Actually Doing

Setting those caveats aside, the documented capabilities represent genuine advances over prior academic BCI systems. BrainGate at Brown University and the University of Pittsburgh's team (working with patient Ian Burkhart) achieved comparable cursor results — but in research settings requiring carts of external recording hardware and tethered electrode arrays. Neuralink's N1 transmits neural data wirelessly over Bluetooth to a paired receiver, requires no external recording hardware during use, and recharges inductively through the scalp. The daily-use portability is a meaningful clinical step.

Documented capabilities across the PRIME cohort through mid-2025:

• 2D cursor navigation: macOS and Windows cursor control sufficient for web browsing, email, and application use; Neuralink reported Arbaugh's cursor speed exceeded prior BCI participants on standard point-and-click benchmarks
• Online chess: Multi-hour Lichess sessions requiring precise targeting and real-time feedback processing
• X (Twitter) posting: First social media post via decoded neural intent, without any secondary input device — Arbaugh posted under his own handle @noarbaugh
• Apple Vision Pro navigation: Spatial interface control using the implant, bypassing the headset's native eye-tracking and hand-gesture systems
• Sustained multi-hour sessions: Without reported fatigue-related performance degradation — a prerequisite for daily clinical adoption

The Bandwidth Reality

The N1 chip transmits raw neural data at approximately 1 Mbps. That figure is not a useful measure of what a patient can accomplish. The meaningful metric is decoded intent bandwidth: how many bits of intended action per second can be reliably extracted from the neural signal and converted to device commands.

Current BCI systems, including Neuralink's, operate in the range of 1–4 bits per second of decoded semantic intent for discrete selection tasks. For cursor control, this translates to roughly 1–2 pointing actions per second at usable accuracy. The best published text-entry rate from an academic BCI is approximately 90 characters per minute, from BrainGate's landmark 2021 handwriting-BCI study (Willett et al., Nature, May 2021), using imagined pen strokes decoded from motor cortex signals. A 2023 follow-on study from Frank Willett's group at Stanford decoded imagined speech at approximately 60–80 words per minute in early data — faster, but still well below natural conversational speech at roughly 150 words per minute.

The gap is not an engineering oversight. The pyramidal tract — the main descending motor pathway — carries an estimated tens of thousands of bits per second of motor intent. Neuralink's 1,024 electrodes sample a fraction of the roughly 100 billion neurons in the human brain. Decoded bandwidth will improve as electrode counts scale and decoding algorithms mature, but it will do so incrementally. The product name "Neuralink Telepathy" is a marketing choice. Broadband neural interfacing at a scale useful for cognitive augmentation remains an open research problem, not an announced engineering roadmap item.

Synchron: The Endovascular Alternative

Synchron's Stentrode approaches motor cortex from a different anatomical direction entirely. A catheter threads through the jugular vein to the superior sagittal sinus — a large venous channel running directly above the motor cortex — where the device expands like a cardiac stent and makes contact with the vessel wall. Neural signals pass through the dura and cortical tissue; the electrodes detect them from outside without penetrating either structure. No craniotomy is required.

Synchron's COMMAND trial enrolled patients with severe ALS in Australia and the United States. The company's first US implant was performed at Mount Sinai Health System in New York in July 2022. Patients have used the Stentrode to control an iPad, send texts and emails, operate smart home devices, and access telehealth apps via decoded motor-intent signals captured from within the vein. Synchron holds FDA Breakthrough Device Designation and reported no serious device-related adverse events in published COMMAND safety data through mid-2025. Investors in the company include Bill Gates, Jeff Bezos, and Khosla Ventures.

The functional tradeoff is real. The Stentrode carries approximately 16 electrodes versus Neuralink's 1,024. Signal quality through vascular tissue is lower than signal from penetrating intraparenchymal electrodes, and the range of decodable commands is narrower. Synchron's clinical thesis is deliberately scoped: restore communication for locked-in patients, not broad motor control. For that constrained goal, the endovascular approach offers a regulatory and risk profile that may prove easier to navigate — no craniotomy, no cortical penetration, and a device that is in principle catheter-retrievable, which no current penetrating BCI device can claim.

Precision Neuroscience: Surface Without Penetration

Precision Neuroscience, co-founded by Ben Rapoport (a Neuralink alumnus) and CEO Michael Mager, uses a third architecture. The Layer 7 Cortical Interface is a thin-film electrocorticography (ECoG) array — 0.1 mm thick — that lays on the brain's surface rather than penetrating it. The current clinical strategy deploys Layer 7 as an add-on during other planned neurosurgeries: epilepsy resections, tumor removals. Surgeons place the array on the exposed cortex for intraoperative recordings to establish safety and signal-quality data, then remove it before closing, building the evidence base needed to pursue a standalone chronic-implant FDA indication.

By early 2025, Precision Neuroscience had completed intraoperative recordings in more than a dozen patients across partner surgical centers. ECoG signals are less spatially precise than single-unit recordings from penetrating arrays — individual neuron spikes are blurred by dura and cortical tissue — but the tissue response is substantially lower. There is no electrode track through cortical layers, no micromotion-driven thread retraction, and no chronic foreign-body inflammation from intraparenchymal hardware. The company has not yet filed for FDA IDE approval for a permanent chronic implant; that milestone awaits the safety record being built now.

Safety and Revision Rates: The Honest Numbers

Each approach carries a distinct risk profile, and honest comparison requires acknowledging significant data gaps across all three platforms:

• Neuralink (penetrating): Thread retraction in Patient 1 is the highest-profile adverse event in current BCI clinical history. Chronic microelectrode signal degradation is documented across decades of academic BCI literature and is not specific to Neuralink's device. Long-term human data beyond 18 months does not exist for the N1 implant. No peer-reviewed surgical outcomes analysis has been published.
• Synchron (endovascular): Standard endovascular procedural risks — thrombosis, vessel injury, catheter-access infection. COMMAND published data through mid-2025 reported no device-related serious adverse events. The device is theoretically catheter-retrievable, a meaningful safety advantage that has not yet been clinically tested in an actual removal scenario.
• Precision Neuroscience (ECoG): Intraoperative-only risk profile for current studies; craniotomy is still required. Chronic implant safety data does not yet exist.

The Actual Trajectory

The realistic near-term horizon for BCIs is not telepathy or cognitive augmentation. It is reliable, wireless motor restoration for people with severe paralysis. The chess games, X posts, and Vision Pro demos represent the current upper bound of demonstrated capability, achieved under favorable conditions with significant engineering support and motivated, otherwise-healthy patients. Broad clinical deployment requires resolving thread retention at the five-year scale, reducing surgical burden below what any current platform requires, and building reimbursement pathways that do not exist for any BCI device today.

The plausible five-year outcome is one or more devices cleared by the FDA for narrow indications — ALS, high cervical spinal cord injury — providing reliable cursor control and text entry at speeds competitive with current eye-tracking and switch-scanning alternatives. That would be a genuinely important medical advance for patients who have lost motor function. It is also a considerably more modest claim than the coverage surrounding each PRIME patient milestone. The distance between Noland Arbaugh playing chess on Lichess and the broadband neural link suggested by Neuralink's marketing is not a gap that incremental electrode scaling closes quickly.