Neuralink Robot-Arm Demo Shown, CONVOY Results Pending

Neuralink has advanced from planning robot-arm experiments to showing a participant use its implant to control an assistive arm for everyday actions. A January 28, 2026 Neuralink update identifies the participant as Nick and documents feeding, serving a drink, gesturing and scratching an itch.

The demonstration changes the status of the story, but not its central limitation: CONVOY remains a small early feasibility study rather than a completed trial or an available medical treatment. Its public registry has no aggregate results for assessing how reliably the arm works, whether performance persists or how the investigational system affects safety and quality of life.

What the demonstration establishes

The footage provides evidence that at least one implanted participant progressed from controlling a computer interface to moving a physical assistive device. Nick had been unable to move his limbs for four years, and the recorded tasks involved directing the arm around his body and nearby objects rather than moving a cursor across a screen.

That is a consequential technical step. A cursor error generally remains on a display, whereas an unintended arm movement can strike a person, spill a drink or miss an object. Useful physical control therefore depends not only on decoding intent, but also on stable movement, safe interaction with the environment and coordination between the brain-computer interface and the robotic hardware.

The demonstration nevertheless supports only a narrow conclusion. It confirms that selected actions occurred; it does not reveal success rates, failed attempts, session duration, calibration time or how much automated assistance the arm supplied. Without those details, the footage cannot establish that the same performance is repeatable across days or participants.

What CONVOY is designed to measure

The official CONVOY study record describes a prospective, non-randomized, open-label, single-group feasibility study sponsored by Neuralink. It lists an estimated enrollment of three participants, invitation-only recruitment and one location at Barrow Neurological Institute in Phoenix. Participants must already be enrolled in the PRIME study and have an N1 Implant.

The primary outcome concerns whether participants can modulate brain activity to control an assistive device during the three months after first use. Secondary measures include device-related adverse events, quality of life and suitability for assistive technology, while follow-up may continue for up to 72 months.

The registry gives November 25, 2024 as the actual study start and May 25, 2031 as the estimated completion date. The record was last verified in June 2025 and contains no posted results. The participant demonstration is therefore evidence of task performance, not a finding that CONVOY has met its predefined safety, consistency or quality-of-life outcomes.

An estimated enrollment of three is appropriate for exploratory work but sharply limits generalization. Such a study can uncover practical obstacles and provide early safety observations; it is not designed to establish effectiveness for the wider population of people with paralysis, spinal-cord injury or amyotrophic lateral sclerosis.

How the implant controls an external arm

The N1 Implant records neural activity associated with intended hand and arm movement. Decoding software translates patterns in those signals into commands for an external device, bypassing the damaged biological pathway between the brain and the participant’s limbs.

This process is more accurately understood as decoded neural control than as an arm responding magically to a thought. The participant and software need a usable mapping between recorded activity and intended movement. Performance can depend on calibration, practice, signal quality, the robotic arm’s supported movements and any assistance built into its control system.

The external arm also does not restore movement to the participant’s own muscles. Approaches that stimulate muscles, peripheral nerves or the spinal cord address a different clinical and engineering problem. CONVOY instead tests whether an implanted interface can provide control over separate assistive equipment.

Why a video is not a clinical result

A selected video can demonstrate feasibility but cannot show the distribution of outcomes. Reliability requires measurements across repeated sessions, while safety assessment requires systematic reporting of adverse events and circumstances in which the device did not behave as intended.

Context about the robotic system matters as well. Reaching for a cup may combine direct neural commands with software constraints, collision avoidance, stabilized trajectories or automated grasping. The public material does not provide enough detail to separate the participant’s decoded control from every function performed by the arm itself.

There is also no public evidence that the Neuralink-controlled arm is available for routine use outside the research setting. Enrollment is by invitation, depends on participation in PRIME and requires an existing implant with adequate performance. A clinical-trial listing and a successful demonstration should not be interpreted as regulatory clearance or commercial availability.

Brain-controlled robotic arms are not new

Neuralink is testing its own implant and assistive-device configuration, but it is not the first research group to achieve human neural control of a robotic limb. A 2012 peer-reviewed BrainGate study involved two people with long-standing tetraplegia who directed three-dimensional reaching and grasping movements using motor-cortex signals recorded by implanted arrays; one participant used a robotic arm to drink from a bottle.

That earlier result established the scientific feasibility of useful robotic-arm control years before CONVOY. Neuralink’s relevant contribution is the application of its own wireless implant, decoding software and external-arm setup in a new human protocol. The unanswered question is whether this configuration can deliver control that is safe, consistent and useful over extended periods.

What remains unknown

The public record provides no robotic-arm task success rate, comparative performance data, typical session length or participant-level analysis of adverse events. It also does not quantify how much independence the system adds relative to other assistive controls available to each participant.

The evidence consequently supports a precise update: Neuralink has progressed from an approved experiment to a documented participant demonstration involving practical object-related actions. Whether that milestone becomes a persuasive clinical result will depend on reported measurements of reliability, safety and daily utility across the study—not on the existence of one successful recording.

Neuralink Shows Robot-Arm Control, but CONVOY Has No Posted Results