BCI & Neuroscience

Brain-computer interfaces by mid-2026 have crossed from "does it work" to "which application reaches approval first, and does it work durably outside the lab." The invasive clinical layer is now a three-way human race: Neuralink (≥24 patients by May 2026; PRIME/CONVOY cursor + robotic arm, plus new VOICE speech and Blindsight vision programs; $650M raise), Synchron (COMMAND endpoint met, preparing a 2026 pivotal trial toward the first-ever FDA PMA for a permanent implant), and Paradromics (first-in-human Connexus implant June 17 2026 — 421 electrodes, >200 bits/sec preclinical, speech-first). Each occupies a distinct point on the invasiveness ↔ bandwidth ↔ regulatory-progress axes.

The defining research result of the period is Card et al. (Nature Medicine 2026): an ALS/dysarthria participant used a multimodal speech+cursor intracortical BCI >3,800 hours at home over ~2 years with no researchers present (183,060 sentences; 56 wpm; 92% ≥mostly-correct) — the first strong evidence that speech BCI works as a sustained daily assistive device, not just a one-session benchmark. This complements the earlier accuracy milestones (BIT's 10% WER; Stanford inner-speech decoding) and the structural reviews of signal acquisition, clinical applications, and AI decoding (CUHK Shenzhen, Deng et al., VIT Chennai, the Multimodal-BCI Transformers review).

Frontier — What's Moving Now

• Invasive race is three-way, all in humans — Neuralink (scale + new VOICE/Blindsight), Synchron (first PMA pathway), Paradromics (first Connexus implant, >200 bps).
• Speech BCI leaves the lab — Card et al.: >3,800 hrs unsupervised at-home use, 56 wpm, 92% correct.
• Synchron → first BCI PMA — COMMAND endpoint met (NCT05035823); 2026 pivotal trial is the gate.
• Neuralink ≥24 patients — CONVOY robotic arm; VOICE speech (FDA Breakthrough); Blindsight vision; $650M raise.
• Sensory neuroprosthetics emerging — Blindsight stimulates visual cortex for the blind (first trials 2026).
• AI for EEG at lab ceiling — 99-100% on constrained tasks; generalizability unsolved; multimodal Transformers as next-gen decoder.
• Closed-loop neurostimulation maturing — Neuropace RNS (epilepsy), adaptive DBS (Parkinson's).

Concept Map

Concepts

Entities

Timeline

See timeline.md for chronological developments (2004 through 2026).

Research Frontier

See frontier.md for active research directions, breakthroughs, and knowledge gaps.

Sources