Kurzweil Scorecard: Scanning the Brain From Inside the Bloodstream, Just Not With Nanobots

In 2005, Ray Kurzweil devoted an entire section of The Singularity Is Near to a specific claim about how we would read the living human brain. It would not be with bigger MRI scanners. It would be with swarms of blood-cell-size devices circulating through the brain’s capillaries, listening to individual neurons and beaming the signals out. He was specific about who would build them — nanotechnologist Robert Freitas — and specific about where they would sit: the spiral artery for hearing, the semicircular canals for balance, the motor neurons for limb position, the olfactory bulb for smell.

It is now 2026. Across our 9.3 million US patents, two mention the word “nanobot.” Neither is in the brain.

And yet, if you squint, Kurzweil was right about something important: it is now possible to read motor cortex activity by snaking a device through the jugular vein and parking it in a cerebral venous sinus, no craniotomy required. Ten patients walk around with one today. It is just not a nanobot. It is a stent.

The predictions

Batch 19 bundles twelve of Kurzweil’s “Scanning Using Nanobots” claims from chapter 4, nearly all pegged to “the 2020s.” They fit into three clusters:

1. Nanobots become viable and scan the brain. “By the 2020s nanobot technology will be viable, and brain scanning will be one of its prominent applications” (ch. “Scanning Using Nanobots”). Once that happens, “billions of nanobots will allow highly sensitive, very high-resolution, noninvasive real-time scanning of a living whole brain.”
2. Freitas-designed monitors for each sense. Specific nanodevices parked near the organ of Corti (hearing), at afferent nerve endings of hair cells in the semicircular canals (balance), at motor neurons (limb position and control), and at skin and mucosal receptors (smell, taste, pain, temperature).
3. BBB workarounds. “An intermediate nanobot strategy will keep the device in the bloodstream while projecting a robotic arm under 20 nanometers wide through the BBB up to about 50 microns to reach nearby neurons,” and, failing that, ventricle injection so nanobots can follow neuron migration paths into tissue.

In The Singularity Is Nearer (2024), Kurzweil conceded the timeline had slipped. He now writes that “at some point in the 2030s we will reach this goal using microscopic devices called nanobots,” adding that this “won’t require some kind of sci-fi brain surgery” because “we’ll be able to send nanobots into the brain noninvasively through the capillaries.” Same mechanism. Plus-ten years.

Where we actually are

The Stentrode is the closest thing that works. In 2023, JAMA Neurology published a four-patient safety study of Synchron’s Stentrode, a self-expanding nitinol mesh of 16 electrodes delivered via catheter through the jugular vein into the superior sagittal sinus, draped against the motor cortex from inside the vein (Mitchell et al., doi 10.1001/jamaneurol.2022.4847, 143 citations). No neurological safety events at twelve months. In October 2024, the COMMAND trial reported 100% accurate deployment across six patients with no neurologic safety events and reliable motor-intent decoding. By late 2025, Synchron had ten patients implanted, raised $200M, and was in pivotal-trial talks with FDA.

The Stentrode is the bloodstream-delivered, non-craniotomy brain reader Kurzweil wanted. It is also about 1,000 times larger than a red blood cell and reads from within veins, not capillaries. Call that wrong-mechanism but right-direction: the route of entry is exactly what he described, and the outcome (reading motor intent in patients with paralysis without opening the skull) is a direct hit.

US 12,168,135, granted December 2024, makes the pattern explicit. Claim 1 recites “detecting, using a first electrode array, an electrophysiological signal of a subject, wherein the first electrode array is coupled to a first endovascular carrier implanted within the subject in at least one of a superior sagittal sinus, an inferior sagittal sinus, a sigmoid sinus, a transverse sinus, and a straight sinus,” and then stimulating via a second endovascular carrier in cortical drainage veins. A closed-loop endovascular pacemaker for epilepsy. US 11,491,323 extends the same architecture to a visual prosthesis — electrodes placed “into the ventricular system or cerebral venous sinuses” to stimulate cortex and restore vision. Ventricular placement was Kurzweil’s own backup plan when the BBB proved too difficult.

Neuralink took the surgical route and is scaling. The 2019 white paper (Musk et al., doi 10.1101/703801, 163 citations) described 3,072 electrodes across 96 flexible threads inserted by a neurosurgical robot. By late 2025, Neuralink had 13 humans implanted across the US, Canada, and UK. Brown’s neurograin program has demonstrated salt-grain-scale wireless sensors that could scale to ~770 per animal in current configurations, with a target of thousands. None of this is bloodborne. All of it requires craniotomy or dura penetration. And all of it records orders of magnitude fewer neurons than the “billions of nanobots” Kurzweil invoked.

Nanobots have started crossing the BBB — for drug delivery, not scanning. A 2026 paper in Wiley (work 7125978334) describes “an allosteric DNA nanorobot for targeted glioma therapy, engineered via DNA origami for acid-triggered doxorubicin delivery,” with in vivo evidence that the nanorobots cross the BBB and accumulate in glioma tissue. A 2026 review in Annals of Medicine and Surgery frames nanorobots crossing the BBB as “the next frontier” of targeted chemotherapy, citing neutrophil-based biohybrid motors that autonomously traverse the BBB along inflammatory gradients to clear residual glioma cells. A 2023 paper (work 4321085913, 43 citations) showed reconfigurable paramagnetic nanoparticle swarms with upstream motility in blood-flow conditions — the kind of propulsion the book skips over entirely.

What none of these have done is record from a neuron. Drug delivery and sensing are different problems. A nanorobot dumping doxorubicin into a tumor doesn’t need to preserve millisecond temporal resolution or communicate 1,000 channels of neural data back out. The scanning half of Kurzweil’s vision is much harder than the crossing half, and harder than the drug-delivery problem that is currently driving all the preclinical progress.

Freitas’s specific designs remain theoretical. Wikipedia summarizes the state of the respirocyte bluntly: “Current technology is not sufficient to build a respirocyte due to considerations of power, atomic-scale manipulation, immune reaction or toxicity, computation and communication.” That verdict extends to every Freitas design on the batch list. No one is sensing auditory signals from inside the spiral artery. No one has a vestibular nanomonitor at hair-cell afferents. No one has olfactory or gustatory nanosensors. Cochlear implants and peripheral-nerve prostheses exist and work, but they are surgically placed macroscale devices.

The deeper goal is being met by an unexpected substitute. Alex Huth’s lab at UT Austin showed in Nature Neuroscience (2023, doi 10.1038/s41593-023-01304-9) that fMRI plus a large language model can reconstruct continuous speech-like sequences from cortical semantic representations — no electrodes, no surgery, no nanobots. In 2025, Communications Biology published an auto-regressive extension capable of reconstructing 10-minute language stimuli. This is not neuron-level, and it inherits fMRI’s blood-flow lag. But it is noninvasive thought-to-text decoding, delivered by the AI wave Kurzweil did predict, using scanners he declared inadequate. The outcome is arriving from a completely different direction than the one the book expected.

The scorecard

What Kurzweil missed (and what he nailed)

He nailed the route. Getting into the brain through the vasculature without a craniotomy is exactly what the Stentrode does, and the patent record shows others converging on the same idea from the vein side. He nailed the demand: patients with ALS and paralysis are already walking around with bloodstream-delivered brain readers, and ten Neuralink patients control computers with their thoughts. The “noninvasive interface to the cortex” he predicted for the 2020s is operational; it just belongs to surgery and interventional radiology, not molecular manufacturing.

What he missed was scale. The book assumed the path from blood-cell-size hypothetical device to functioning molecular-scale neural interface would be paved by the exponential curves that carried computing from vacuum tubes to smartphones. Biology is not cooperating. Mechanosynthesis, atomic-scale fabrication, in-body power, and immune compatibility have all advanced — but nothing like 2^n per two years. Meanwhile, the fields Kurzweil did not foreground — endovascular neurology, DNA origami chemistry, focused-ultrasound BBB opening, and large language models reading fMRI — have each quietly moved the goalposts of his prediction from inside.

If there is a pattern across batch 19, it is that Kurzweil’s ends are being met by someone else’s means. The interesting question for forecasters is no longer whether Freitas was wrong, but whether his specific designs will ever become cheaper than the catheter, the AAV vector, the LLM, and the 1,024-electrode thread that are beating him to the clinic.

Method note

We read the predictions in The Singularity Is Near (2005) and cross-checked them against restatements in The Singularity Is Nearer (2024). We then mined a 9.3M-document US patent corpus and a 357M-paper OpenAlex literature corpus for every relevant keyword — nanorobot, endovascular electrode, neurograin, DNA origami BBB, focused ultrasound, microrobot swarm, two-photon capillary imaging, and each Freitas design — then read the actual claims of the most-cited patents and the abstracts of the highest-citation papers. Web searches filled in clinical-trial counts, funding, and product milestones through April 2026. Scorecard verdicts reflect the evidence found this session, not general impressions.

Sources:
– Mitchell et al., JAMA Neurology (2023)
– Synchron COMMAND early feasibility study summary
– Synchron raises $200M (Nov 2025)
– Neuralink clinical trials update
– Musk et al., Neuralink white paper (2019)
– Allosteric DNA nanorobot for glioma (2026)
– Nanorobots crossing the BBB review (2026)
– Huth lab semantic decoding, Nature Neuroscience (2023)
– Brown neurograin wireless BCI (2024)
– Respirocyte state-of-the-art summary