03 — Brain–Computer Interfaces (BCI): Neuralink & the 2026 scene

Part of the /workspace/health/biohacking/ landscape overview. Lens: BCIs as a possible bridge toward substrate independence / mind-uploading ("Friendship is Optimal"), and as a component of "biopunk 2037." Author cares about real 2026 status, demonstrated-vs-hype, and what (if anything) updates the uploading timeline. Research date: 2026-05-31. Confidence tiers C1–C5 per /workspace/RESEARCH-BEST-PRACTICES.md.

TL;DR (read this first)

1. Invasive BCI is real and accelerating — for restoring function, not for uploading. Neuralink has ~21 humans implanted across 4 countries (early 2026); the field's best demonstrated outputs are cursor control at ~8–10 bits/sec, robotic-arm self-feeding, and speech decoding at ~62 words/min / ~75% accuracy on a 125k-word vocabulary. These are output/read interfaces for paralyzed patients.
2. The field is multi-player and well-funded. Neuralink ($9B valuation, >$1.3B raised), Synchron (endovascular, $345M raised, Gates+Bezos), Precision (FDA-cleared 1,024-electrode surface array), Paradromics (1,600+ channels, first human recording 2025), Science Corp (Max Hodak, $490M, biohybrid + PRIMA retinal implant), Blackrock Neurotech (Utah array, 20-yr human track record).
3. Channel counts are the binding constraint and they are catastrophically short of uploading. State-of-the-art human implants read ~1,000–1,600 neurons out of ~86 billion. That is ~7 orders of magnitude. BCIs read/write a thin I/O channel; they do not, and on any current trajectory cannot, copy a brain. "BCI = uploading" is essentially false as stated.
4. Relevance to uploading is indirect: BCIs advance neural-recording hardware, decoding ML, and regulatory/surgical pathways — useful adjacent tech — but uploading needs connectomics + preservation + emulation (see 04-wbe-uploading.md), not better cursors.
5. Skeptic take: genuine medical breakthroughs are wrapped in heavy CEO hype (Musk's "telepathy," "save/replay memories"); peer-reviewed literature explicitly flags "upload/download your mind" as a misconception. Animal-welfare controversy (≈1,500 animals died in Neuralink testing) and a thread- retraction failure in patient #1 are real caveats.
6. Non-invasive/consumer (EEG, fNIRS, ultrasound) is low-bandwidth and mostly wellness/research-grade — essentially zero direct relevance to uploading.

Part A — Invasive BCI companies

Neuralink (N1 "Telepathy" implant)

Claim: ~21 human participants ("Neuralnauts") enrolled in clinical trials worldwide as of early 2026. - Confidence: C2 (company self-report via "Two Years of Telepathy," Feb 2026; corroborated by secondary press) - Sources: https://neuralink.com/updates/two-years-of-telepathy/ ; https://teslanorth.com/2026/01/28/from-paralysis-to-neuroscience-how-21-people-are-using-neuralink-in-2026/ ; https://www.technology.org/2026/01/29/neuralink-brain-chips-connect-21-minds-to-machines/ - Notes: A TechCrunch/SiliconANGLE report (mid-2025) cited "12 implant recipients across four countries by late 2025," so 21 by early 2026 is a fast ramp — plausible but treat the exact figure as company-sourced. https://techcrunch.com/2025/05/28/elon-musks-neuralink-raises-600m-at-9b-valuation/

Device. N1 = 1,024 electrodes on 64 flexible polymer threads, inserted by the R1 robot into motor cortex; fully wireless. Trial = PRIME study (NCT06429735), tetraplegia from cervical SCI or ALS; primary completion 2026, full study to 2031. - Confidence: C1 (clinicaltrials.gov registry + company) - Sources: https://clinicaltrials.gov/study/NCT06429735 ; https://neuralink.com/updates/two-years-of-telepathy/

Demonstrated capabilities (real): - Cursor control: participant "Nick" reportedly >10 bits/sec in first week (matches/exceeds able-bodied mouse use, ~8–10 BPS). Demonstrated. - Robotic-arm control for self-feeding. Demonstrated (single participant, company video). - Typing via virtual keyboard up to ~40 wpm; first patient Noland Arbaugh uses device ~10 hrs/day for study/gaming/scheduling. Demonstrated. - Sources: https://teslanorth.com/2026/01/28/... ; https://mezha.ua/en/news/neuralink-telepathy-2-years-of-tests-308110/ ; https://www.elonx.net/neuralink-in-2024-the-year-of-noland-arbaugh-the-first-human-user-of-the-n1-chip/amp/ - Confidence: C2 (company demos + press; not all peer-reviewed)

Aspirational / in-progress: - VOICE trial — real-time speech restoration by reading speech-motor areas. Announced/early. - Blindsight — visual-cortex stimulation implant for blindness. FDA Breakthrough Device designation Sept 2024 (NOT approval; no human implant yet as of mid-2026, "expected 2026"). Musk's "better than natural vision / see in IR/UV" claims are aspirational; IEEE Spectrum notes early resolution will be crude ("Atari graphics") and it won't deliver natural sight. - Sources: https://www.mobihealthnews.com/news/elon-musks-neuralink-device-blindsight-gets-fda-breakthrough-device-designation ; https://spectrum.ieee.org/neuralink-blindsight - Confidence: C2 (designation is C1; capability claims C4) - Next-gen device "~3× capability" planned later in 2026. Aspirational.

Funding: $650M Series E (June 2025) at ~$9B valuation; >$1.29B raised total. Lead: ARK, Sequoia, Founders Fund. - Confidence: C2 — Sources: https://techcrunch.com/2025/05/28/elon-musks-neuralink-raises-600m-at-9b-valuation/ ; https://siliconangle.com/2025/06/02/neuralink-raises-another-650m-reported-9b-pre-money-valuation/

Controversies (skeptic anchor): - Animal welfare: ≈1,500 animals (incl. 280+ sheep/pigs/monkeys) died in testing since 2018 (Reuters, company estimates); USDA opened an investigation Dec 2022; July 2023 USDA found no breaches beyond a self-reported 2019 incident, but PCRM disputed this; ≥1 Animal Welfare Act violation acknowledged (delayed euthanasia of a monkey). Nov 2023: lawmakers asked SEC to probe possible investor deception. - Sources: https://en.wikipedia.org/wiki/Neuralink ; https://awionline.org/awi-quarterly/winter-2023/investigation-uncovers-disturbing-details-neuralink-monkey-study ; https://eandt.theiet.org/2022/12/07/musks-neuralink-investigated-over-death-1500-animals - Confidence: C2 - Thread retraction: ~85% of Noland Arbaugh's threads retracted ~3 months post-surgery (Jan 2024), likely from trapped intracranial air; fixed in software (more sensitive recording algorithm), not re-surgery. A real hardware-durability red flag. - Sources: https://www.nbcnews.com/science/science-news/implant-elon-musks-neuralink-suffers-setback-threads-retract-patients-rcna151443 ; https://futurism.com/neoscope/first-neuralink-patient-trouble-implant - Confidence: C2

Uploading-bridge relevance: Low-direct, moderate-indirect. High channel count (1,024) and a scalable robotic-insertion + flexible-thread platform push recording hardware forward — but it's a read/write interface, not a brain-copying technology. Musk's "merge with AI / preserve memories" framing is not supported by anything demonstrated.

Synchron (Stentrode — endovascular)

Claim: Stentrode implanted in 10 patients (US + Australia) across trials; less invasive — delivered via jugular vein into a vessel over motor cortex (no open-skull surgery). - Confidence: C2 — Sources: https://www.fiercebiotech.com/medtech/synchron-raises-200m-advance-its-brain-computer-interface-paralysis ; https://www.massdevice.com/gates-bezos-synchron-75m-bci/

Funding: $200M Series D (Nov 2025); total ~$345M. Earlier Series C (2022) backed by Bill Gates' Gates Frontier and Jeff Bezos' Bezos Expeditions; Bezos Expeditions returned in Series D. - Confidence: C1/C2 — Sources: https://www.businesswire.com/news/home/20251106150841/en/Synchron-Raises-$200-Million-Series-D-to-Advance-Brain-Computer-Interface-Technology ; https://www.massdevice.com/gates-bezos-synchron-75m-bci/

Demonstrated: FDA-approved COMMAND trial; ALS patient Mark Jackson first to control an iPad by thought. 2025: native BCI integration roadmap with Apple (iPhone/iPad/Vision Pro via Switch Control) and edge-AI demo on NVIDIA Holoscan displayed on Apple Vision Pro. - Sources: https://www.businesswire.com/news/home/20250513927084/en/Synchron-To-Achieve-First-Native-Brain-Computer-Interface-Integration-with-iPhone-iPad-and-Apple-Vision-Pro ; https://www.massdevice.com/synchron-cognitive-ai-model-bci-nvidia/ - Confidence: C2

Aspirational: Chiral™ — "world's first cognitive AI brain foundation model," trained on de-identified neural data with NVIDIA. Heavy on framing ("Cognitive AI"); the actual deployed capability is still assistive control for motor-impaired users. Treat "foundation model of cognition" language as marketing. - Sources: https://www.fiercebiotech.com/medtech/synchron-and-nvidia-set-sights-ai-model-trained-direct-brain-activity - Confidence: C3 (claim exists; capability unproven)

Skeptic / tradeoff: Endovascular = far safer/scalable (no craniotomy) but much lower spatial resolution and channel count than penetrating arrays (Stentrode is ~16 electrodes) — so it tops out at coarse control. The safety-vs-bandwidth tradeoff is the central tension in the field.

Uploading relevance: Very low. Lowest-bandwidth of the serious invasive players; valuable as a commercialization/regulatory pathfinder, not as a recording-density advance.

Precision Neuroscience (Layer 7 Cortical Interface)

Claim: Surface (non-penetrating) microelectrode array, 1,024 electrodes, thinner than a human hair, conforms to cortical surface. FDA 510(k) clearance April 2025 for recording/monitoring/stimulating for up to 30 days — reportedly the first BCI cleared for that duration in a clinical setting. - Confidence: C1/C2 — Sources: https://www.cnbc.com/2025/04/17/brain-implant-cleared-fda-musk-neuralink-rival-precision-neuroscience.html ; https://en.wikipedia.org/wiki/Precision_Neuroscience ; https://www.fiercebiotech.com/medtech/fda-clears-precision-neurosciences-minimally-invasive-brain-computer-interface-implant - Note: 510(k) clearance ≠ approval of a consumer/long-term implant; it's a regulated medical-device clearance for short-duration use. Don't overstate.

Key person: Benjamin (Ben) Rapoport — neurosurgeon (Mount Sinai), Neuralink co-founder who left to co-found Precision (2021) explicitly betting on a minimally invasive, reversible surface array vs. penetrating threads. Good signal that even ex-Neuralink insiders see penetration-depth as a liability. - Source: https://en.wikipedia.org/wiki/Precision_Neuroscience — Confidence: C2

Skeptic / tradeoff: Surface arrays read cortical-surface field potentials, not single neurons in deep layers → lower resolution than penetrating arrays but safer and removable. Same safety-vs-bandwidth tradeoff.

Uploading relevance: Low-direct. The 30-day-implant clearance and reversibility matter for clinical scale, not for the deep, dense, whole-brain recording uploading would require.

Paradromics (Connexus)

Claim: First-in-human recording June 2025 (20-min temporary placement during epilepsy surgery, Univ. of Michigan). >1,600 channels (421 microwire electrodes/module × up to 4 modules = 1,684 intracortical channels). FDA IDE approval Nov 2025 for the Connect-One early-feasibility study — first IDE for speech restoration with a fully implantable BCI. - Confidence: C2 (two sources) — Sources: https://www.paradromics.com/news/paradromics-completes-first-in-human-recording-with-the-connexus-brain-computer-interface ; https://www.michiganmedicine.org/news-release/university-michigan-team-leads-first-human-recording-new-wireless-brain-computer-interface ; https://www.massdevice.com/paradromics-fda-green-light-bci-study/ - Note: First implant was temporary proof-of-concept, not a chronic implant. Highest channel count among the human-stage players.

Uploading relevance: Of the invasive players, Paradromics + Neuralink push raw channel count hardest — the dimension most relevant to "more neurons read." Still ~7 orders of magnitude short of whole-brain.

Blackrock Neurotech (Utah array)

Claim: Utah array (≤100 channels/device) — the longest human track record in BCI: implanted in humans since 2004; >30,000 cumulative in-patient BCI research-days (~20× any other implantable BCI). Individual arrays have run 987 and 182 days in participants. Developing MoveAgain clinical system (FDA Breakthrough Designation 2021). - Confidence: C2 — Sources: https://blackrockneurotech.com/insights/record-set-implantable-brain-tech/ ; https://www.medicaldesignandoutsourcing.com/utah-array-brain-computer-interface-blackrock-neurotech/ - Note: The Utah array is the workhorse behind most academic BCI milestones (BrainGate, UCSF/Stanford speech work). Mature, lower channel count than newer entrants, but the reliability/longevity benchmark.

Uploading relevance: Low-direct; high as the proof that chronic human cortical recording works for years.

Science Corp (Max Hodak — biohybrid + PRIMA)

Claim: Founded 2021 by Max Hodak (Neuralink co-founder and ex-president). Two thrusts: 1. PRIMA photovoltaic retinal implant for late-stage macular degeneration — completed pivotal trials, reportedly >80% of patients with meaningful visual-acuity improvement; FDA Breakthrough status 2023; pursuing FDA + EU approval. 2. Biohybrid BCI ("Biostar"-type) — uses living neurons grown to interface with the brain instead of rigid wires, aiming to sidestep foreign-body damage and scale connection count biologically. - Funding: $230M Series C (early 2026); ~$490M total raised. Investors: Lightspeed, Khosla, Y Combinator, IQT, Quiet Capital. - Confidence: C2 — Sources: https://science.xyz/news/series-c/ ; https://techcrunch.com/2026/03/05/science-corp-closes-230m-round-as-it-pushes-to-get-its-brain-implant-to-patients/ ; https://www.mobihealthnews.com/news/neural-engineering-company-science-corporation-garners-230m - Note: PRIMA is the most concretely delivered product of any company here (restored functional reading vision). The biohybrid BCI is the most speculative, but conceptually the most interesting for high-channel-count futures — using biology to multiply connections is the only paradigm that even gestures at the scale problem.

Uploading relevance: Biohybrid approach is the most architecturally novel; if living-neuron interfaces scaled to millions of connections it would matter — but that is years-to-decades out and unproven in humans.

China — the #2 BCI ecosystem (the corpus's biggest geographic gap)

Claim: China is now arguably the second BCI ecosystem after the US, on a deliberately semi-/minimally-invasive track (electrodes that don't penetrate the dura — lower bandwidth than Neuralink's penetrating threads, but lower infection/scarring risk). - NEO (Tsinghua University / Neuracle, team led by Hong Bo): a wireless, minimally invasive (epidural, dura-sparing) implant. First human implant 2023 (Xuanwu Hospital); scaled to larger clinical trials in 2025 (target ~50 patients), and China's NMPA cleared NEO for commercial/medical use (reported Apr 2026) — framed in Chinese press as the first BCI approved for use outside strictly controlled trials, ahead of Neuralink on the regulatory-approval axis. Demonstrated (human) + regulatory clearance. C2 · Tsinghua Medicine (NEO in Nature's "Science in 2025"), People's Daily, Apr 2026, Yicai Global - Beinao-1 / Beinao No.1 (CIBR — Chinese Institute for Brain Research + NeuCyber / Beijing Xinzhida Neurotechnology): a semi-invasive brain chip implanted in 3 patients in early 2025, with a stated target of ~50 patients by 2026 — China's state-backed answer to Neuralink, with a national BCI initiative behind it. Demonstrated (human), early. C2 · Yicai Global - Skeptic take: semi-invasive = lower channel count / bandwidth than penetrating arrays, so on raw neural-read capability these trail Neuralink/BrainGate; but on deployment pace and regulatory approval China is moving fast and is a genuine #2. For the uploading path (Part D) the bandwidth ceiling matters more than the approval headline — semi-invasive BCIs are even further from connectome-scale read than penetrating arrays.

Part B — Academic milestones (the real capability frontier)

Most headline BCI capability is academic, usually on Blackrock Utah arrays under the BrainGate consortium (Hochberg/Donoghue; Brown/MGH/Stanford/UC Davis), not from the startups.

BrainGate consortium. 14 participants enrolled 2004–2019; 20 arrays; 2,319 recording sessions; 20 years of data. Arrays recorded usable spikes on ~35.6% of electrodes with only ~7% decline over up to 7.6 years — landmark longevity evidence. - Confidence: C1 — Source: https://www.medrxiv.org/content/10.1101/2025.07.02.25330310v1 (also PMC12236888)

Handwriting decoding (Willett/Shenoy, Stanford, Nature 2021). Decoded imagined handwriting → text at 90 characters/min, 94.1% raw / >99% with autocorrect — fastest BCI typing then reported. - Confidence: C1 — Sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC8163299/ ; https://www.brown.edu/news/2021-05-12/handwriting

Speech neuroprosthesis — Stanford/BrainGate (Willett et al., Nature, Aug 2023). Decoded attempted speech at 62 wpm, 9.1% word-error-rate on 50 words / 23.8% WER on a 125,000-word vocabulary — 3.4× the prior speed record; first large-vocabulary demonstration. (Natural conversation ≈160 wpm.) - Confidence: C1 — Sources: https://www.nature.com/articles/s41586-023-06377-x ; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10468393/

Speech — UCSF Chang lab (with UC Berkeley, Nature Neuroscience, Mar 2025). "Brain-to-voice" streaming neuroprosthesis: near-real-time synthesis of a stroke patient's intended speech (digital vocal tract, ~no detectable delay). Earlier Chang work decoded ~78 words/min and drove an avatar. - Confidence: C1/C2 — Sources: https://www.nih.gov/news-events/nih-research-matters/brain-computer-interface-restores-natural-speech-after-paralysis ; https://magazine.ucsf.edu/speaking-again-little-help-ai ; https://www.nature.com/articles/s41586-023-06443-4

Inner-speech decoding (Kunz/Willett, Stanford, Cell, Aug 2025). Decoded imagined (not attempted) speech from motor cortex in 4 participants — up to 74% accuracy on a 125,000-word vocabulary. Inner speech gives weaker signals than attempted speech but removes the fatigue of physically attempting to speak. - Confidence: C1 — Sources: https://www.cell.com/cell/fulltext/S0092-8674(25)00681-6 ; https://news.stanford.edu/stories/2025/08/study-inner-speech-decoding-device-patients-paralysis - Note for the uploading lens: This is the closest anything gets to "reading thoughts," but it's decoding the motor representation of words a person deliberately rehearses — not reading arbitrary cognition, memory, or qualia.

Brain–spine interface / "digital bridge" — Courtine & Bloch (EPFL/CHUV/NeuroRestore + ONWARD Medical), Nature, May 24 2023. Two cortical implants read movement intention from the sensorimotor cortex and wirelessly drive epidural electrical stimulation of the lumbosacral spinal cord — re-establishing the broken brain↔spine link so a chronically paralyzed man (Gert-Jan Oskam, SCI from a cycling accident) could stand, walk naturally, and climb stairs by thought. Strikingly, neurological recovery persisted even with the system switched off, suggesting the stimulation promoted new nerve-fiber growth/reorganization. ONWARD Medical is commercializing the spinal-stimulation side (ARC therapy; FDA Breakthrough designations). Demonstrated in a human. C1 — Sources: https://www.nature.com/articles/s41586-023-06094-5 ; https://pubmed.ncbi.nlm.nih.gov/37225984/ ; https://www.neurorestore.swiss/press-1/bci2023 ; https://www.science.org/content/article/paralyzed-man-walks-naturally-thanks-wireless-bridge-brain-spine - Note for the nerve-reconnection lens (biopunk-2037): This is the strongest real, demonstrated-in-human evidence for the author's "nerve reconnection" worldbuilding element. It does not physically re-fuse a severed nerve — it bypasses the lesion electronically (brain → computer → spinal stimulator) and, as a bonus, appears to encourage native regrowth. So "nerve reconnection" is functionally demonstrated via a digital bridge, while direct biological re-fusion of an arbitrary severed nerve at scale remains unsolved. See doc 08. This corrects an earlier framing in the corpus that called nerve reconnection "unaddressed."

Part C — Non-invasive / consumer neuro

Low risk, low bandwidth, mostly wellness/research/control-signal — negligible direct uploading relevance.

OpenBCI / Galea. CEO Conor Russomanno. Open-source biosensing. Galea Neon (with Pupil Labs): first untethered all-in-one headset fusing EEG + EOG + EMG + EDA + PPG + eye-tracking for VR/research. Research/developer tool, not a control BCI. - Confidence: C2 — Sources: https://openbci.com/community/openbci-and-pupil-labs-partnership-galea-neon/ ; https://www.media.mit.edu/posts/galea/

Kernel (fNIRS). Bryan Johnson's former co (he invested ~$64M; CEO now Ryan Field). Still operating in 2026 (no shutdown). Pivoted from selling Kernel Flow helmets toward brain-data + NeuroAI services: Brain Age Tracker / Cognition Tracker piloted in clinics, dementia-detection and depression-treatment-aid work. fNIRS = blood-flow proxy, low temporal resolution. - Confidence: C2 — Sources: https://en.wikipedia.org/wiki/Kernel_(neurotechnology_company) ; https://www.kernel.com/about

Forest Neurotech (ultrasound). Non-profit Focused Research Organization. Licenses Butterfly Network's ultrasound-on-a-chip to build a minimally invasive functional-ultrasound (fUSI) + neuromodulation BCI ("Forest 1," key-fob sized). ~$20M joint effort. Ultrasound can image deep-brain blood-flow and modulate neurons non-electrically — interesting middle ground, still research-stage. - Confidence: C2 — Sources: https://forestneurotech.org/ ; https://spectrum.ieee.org/bci-ultrasound ; https://www.massdevice.com/butterfly-network-brings-ultrasound-bci-space/

Consumer EEG (Emotiv, Muse, Neurosity). - Neurosity Crown — 8 EEG channels, focus-tracking, developer API; notably ships a native MCP server (real-time brain data to LLMs like Claude/ChatGPT) — gimmick-adjacent but a sign of the AI tie-in trend. - Emotiv EPOC X — packaged research headset; raw-data access gated behind paid developer license + royalties. - Muse (S Athena, Mar 2025) — first consumer wearable combining EEG + fNIRS; optimized for meditation/sleep, frontal/temporal coverage only. - Confidence: C2/C3 (vendor + review sources) — Sources: https://neurosity.co/guides/neurosity-crown-vs-emotiv-epoc-x ; https://www.myndlift.com/post/2026-guide-top-8-neurofeedback-devices-for-brain-training - Note: Vendor/affiliate-heavy sourcing; capability claims for consumer EEG are routinely overstated.

EEG limitations (skeptic anchor). Scalp EEG: poor spatial resolution (cm-scale, blurred by skull), artifact-prone, no deep-structure access, ~µV signals → suitable for coarse state classification (focus/relaxation, P300 spellers, simple BCI control), not for reading words/thoughts/memories.

The honest gap: what BCIs can and CANNOT do (2026)

CAN (demonstrated): restore cursor/keyboard/robotic-arm control for paralysis (~8–10 BPS); decode attempted or imagined speech to text/voice (~62 wpm, ~25% WER large-vocab); coarse state-classification non-invasively; write crude stimulation patterns (Blindsight goal; retinal PRIMA does this for vision). All as a thin, deliberate input/output channel for motor/communication impairment.

CANNOT: read arbitrary thoughts, memories, or subjective experience; "download" knowledge; copy or back up a mind; record more than a tiny fraction of neurons; sustain dense recording indefinitely (threads retract, arrays degrade). Peer-reviewed literature explicitly lists "upload/download your mind" as a misconception propagated by media and some CEOs. - Sources: https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2024.1391550/full ; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11004276/

The bandwidth/scale argument (the load-bearing point): - Human brain ≈ 86 billion neurons, ≈100 trillion synapses. Confidence: C1 — Sources: https://hms.harvard.edu/news/new-field-neuroscience-aims-map-connections-brain ; https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0310156 - Best human implants read on the order of 10²–10³ neurons (Utah ~100 ch; Neuralink/Precision 1,024 ch; Paradromics ~1,600 ch). That's a recording fraction around 10⁻⁷ to 10⁻⁸ of the brain. - A static structural whole-brain connectome at nanoscale is estimated at ~0.2 to ~370 petabytes depending on model fidelity, and imaging the whole human brain at fly-connectome protocol speed would take an estimated ~17 million years with current methods. Confidence: C2 — Sources: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0310156 ; https://www.sciencedirect.com/science/article/pii/S187775032400139X - These are different problems. A live BCI samples activity from a few thousand neurons through a tiny window; uploading needs the full static wiring + molecular state of every synapse, which is a destructive imaging / connectomics problem (see 04-wbe-uploading.md), not an electrode problem. Even Paradromics' own blog argues "brain speed doesn't limit BCI potential" — i.e., they're optimizing communication bandwidth for control, explicitly not mind-copying. (https://www.paradromics.com/blog/brain-speed-bci-potential)

Why "BCI = uploading" is mostly wrong: scaling electrodes by even 1,000× (a heroic decade-plus engineering feat) still leaves ~4–5 orders of magnitude gap, reads only activity not structure, and a chronic implant cannot non-destructively enumerate 100 trillion synapses. BCIs are an interface; uploading needs a map + a substrate. They are adjacent fields, not the same path.

Relevance to the uploading path (FiO bridge assessment)

Bottom line for the FiO timeline: Nothing in the 2026 BCI scene moves up a realistic mind-uploading timeline. BCIs validate that chronic human neural recording works and that ML can decode motor/speech intent — useful confidence-builders and tool-developers — but the uploading critical path runs through connectomics + preservation + emulation, where BCIs contribute almost nothing. If anything, the explicit "BCI = merge with AI / preserve your mind" hype should be discounted, not used to shorten estimates. The biohybrid paradigm (Science Corp) is the one thread worth watching as a conceptual scaling route, but it's pre-human and years-to-decades out. For "biopunk 2037," BCIs are more relevant via the nerve-reconnection / sensory-restoration angle (cursor/limb/speech/vision restoration is genuinely arriving this decade) than via uploading.

What I couldn't verify / open questions (C5 unknowns)

• Exact current Neuralink participant count. Company says ~21 (early 2026); a mid-2025 press figure said 12 by late 2025. The 21 is company-self-reported and not independently audited; treat ±. Failed to find a registry-level confirmed count. (Search terms tried: "Neuralink participant count 2026 verified clinicaltrials".)
• Whether Blindsight has had any human implant yet — as of searches, still "expected 2026," designation only, no confirmed human implant found.
• VOICE / CONVOY (Neuralink robotic-arm) trial registry details — referenced in press but I did not confirm separate ClinicalTrials.gov registrations / enrollment numbers.
• Synchron Stentrode exact electrode count — widely reported ~16 but I did not pin a primary spec sheet here; flagged as approximate.
• Science Corp biohybrid ("Biostar") human-stage status — PRIMA (retinal) is well documented; the biohybrid BCI's first-human timeline I could not firmly date (one source mentioned "first human brain sensor placement with a Yale neurosurgeon" but I did not verify it occurred).
• Consumer-EEG capability claims are vendor/affiliate-sourced and likely optimistic; channel counts are reliable, "focus/cognition tracking" efficacy is not independently verified here.
• Petabyte connectome estimates vary 1000× across models (0.2–370 PB) — order-of-magnitude only; the argument (≥7 OOM gap, structure≠activity) is robust regardless of which estimate.

Cross-links: uploading/WBE/connectomics → 04-wbe-uploading.md; brain preservation → /workspace/cryonics/; chip/grinder implants (different scene) → /workspace/health/chip-implants/; biohybrid/organ tissue → /workspace/health/organ-sacs/.