Will We Merge With AI? The Real Ladder of the Connected Brain: What Is Already Being Implanted in Operating Rooms — and What Remains a Promise

THE THESIS. Merging with AI will not be an event: it will be a ladder, and today we are on the first rungs. Brain-computer interfaces (BCIs: systems that read the brain's electrical activity and translate it into commands for a machine) are medicine today, not transhumanism. They restore lost functions — moving a cursor, speaking — to people with paralysis or ALS (amyotrophic lateral sclerosis, the disease that destroys motor neurons). They expand no one's mind. Our thesis has three parts: what has been demonstrated in operating rooms is extraordinary and moving faster than most people think; the “merge with AI to survive AI” argument does not survive an engineering analysis; and the battle actually being fought in 2026 is not technological but legal: who governs the data that comes out of your brain.

WHAT IS ALREADY BEING IMPLANTED IN OPERATING ROOMS. Three companies define the real ladder, and they differ by how deeply they invade the brain. Neuralink, the most aggressive, inserts electrode-laden threads into the cortex itself: Elon Musk confirmed 21 implanted patients in February 2026, and unofficial trackers following its trials counted more than 25 by mid-year, with active studies in the US, Canada, the UK and the UAE. Its PRIME trial lets quadriplegic people control computers and robotic arms by thought. Synchron, the cleverest, avoids opening the skull: its Stentrode enters through the jugular vein and deploys inside a blood vessel next to the motor cortex. Its COMMAND study closed 12 months with six patients and zero device-related serious adverse events, and the company raised $200 million in November 2025 to fund the pivotal trial that could make it the first BCI with full FDA commercial approval. It also has native Apple integration: its patients control iPhones and the Vision Pro with their minds. Precision Neuroscience, the middle path, lays a film thinner than a human hair carrying 1,024 electrodes on the surface of the cortex — without penetrating it; in April 2025 its Layer 7 won the first FDA 510(k) clearance for a next-generation BCI (use of up to 30 days) and has been tested in 37 patients, according to the company. The money is following: Morgan Stanley estimates a potential market of about $400 billion in the US alone.

GETTING A VOICE BACK, WORD BY WORD. The most moving advance — and the one our audience cites most — is synthetic speech. At UC Davis, within the BrainGate2 trial, Casey Harrell, a 45-year-old man with ALS, received four microelectrode arrays in the left precentral gyrus (the region that coordinates speech): the system translated his attempted speech with 97% accuracy, the best published result (New England Journal of Medicine, August 2024), and it worked within minutes of being switched on. In August 2025, a Stanford-Emory team went further (in the journal Cell): it decoded inner speech — the words you say to yourself without moving a muscle — with up to 74% accuracy, and added something revealing: a mental password (“chitty chitty bang bang”) that keeps the decoder off and blocked 98% of unintended readouts. When the scientists themselves are already designing mental locks, that tells you where this is heading. Neuralink, for its part, says that in its VOICE trial a participant named Terry, who has bulbar ALS, spoke the words “I love you” through a voice rebuilt by AI from recordings made before his illness; that result comes from the company and its trial trackers, with no peer-reviewed publication yet. The nuance that puts everything in order: these systems read the intention to speak in the motor cortex, trained person by person over weeks. They do not read thoughts.

READING THE MIND WITHOUT SURGERY — AND THE FINE PRINT. What about without an implant? In 2023, Jerry Tang and Alexander Huth (University of Texas at Austin) published a “semantic decoder” in Nature Neuroscience: it combines functional MRI (fMRI, a technique that measures blood flow in the brain) with a language model to reconstruct the gist of what a person hears or imagines. It is a milestone, but its fine print is the real story: it needs 16 hours of voluntary training per person, it recovers general meaning rather than exact words, and if the subject resists — by thinking about something else — the system fails. Cooperation is, today, a technical requirement. And the other half of the “merge” simply does not exist: “writing” to the brain — inserting knowledge or memories — is limited to crude stimulation (rudimentary visual percepts, relief of Parkinson's symptoms). Our read: the asymmetry between reading and writing is this field's hard frontier, and it is measured in decades, not product launches.

THE BANDWIDTH ARGUMENT, UNDER EXAMINATION. The symbiosis idea is old: J.C.R. Licklider formulated it in 1960, before the internet. Musk updated it with his bandwidth argument: we are already cyborgs — we live glued to our phones — but we communicate with machines at thumb speed; a BCI would multiply that channel, and merged with AI we would not be left behind. Ray Kurzweil puts a date on it: nanobots connecting the cortex to the cloud around 2045. It is an elegant thesis with an empirical problem. A Caltech study (in the journal Neuron, December 2024) calculated that conscious thought processes about 10 bits per second, while our senses take in about a billion: the great bottleneck does not appear to be the interface, but cognition itself. For calibration: speaking already transmits about 39 bits per second (Science Advances, 2019), and no current implant sustainably beats a keyboard, let alone the voice. Our read: merging as insurance against AI risk is napkin philosophy, not engineering. No one has explained the mechanism by which a cursor-speed link would supervise an AI operating at datacenter speed. The real, verified use case is something else, and it is immense: restoring autonomy to millions of people with paralysis, stroke or ALS.

NEURORIGHTS: THE LAW ARRIVES BEFORE THE MERGE. While we debate the merge, lawmakers are already regulating the current rung — and almost nobody covers this in Spanish. Chile was the world's pioneer: it amended its constitution in 2021 to protect brain activity, and in 2023 its Supreme Court ordered the company Emotiv to delete the neural data of Senator Guido Girardi, captured by a consumer headset. In the US, Colorado became the first state to classify neural data as sensitive information in April 2024, and California followed with bill SB 1223 in September 2024. The capstone came on November 12, 2025: UNESCO's 194 member states adopted the Recommendation on the Ethics of Neurotechnology, the sector's first global standard. Our final read: this is the rare case in which the law gets ahead of the technology, and that is worth celebrating — because the near-term risk is not implants (few in number, very expensive and FDA-supervised) but consumer wearables that already collect brain signals without medical-device status: once again, de facto adoption arrives before authorization. The implications, in our editorial line: in the short term, honesty — there will be overpromising, privacy incidents and a neural-data market to watch; in the long term, what is already happening in operating rooms is the first rung of the abundance horizon: first restoring voice and movement, later — perhaps, with decades and clinical trials in between — expanding capabilities. The merge, if it comes, will not be a switch flipped by a visionary: it will be a medical ladder, climbed rung by rung, with the FDA and neurorights as the handrail.