non-invasive got real, and the money spread.

Last issue, China crossed the line first. This week the story moved somewhere quieter and bigger. Non-invasive brain-to-text got real, serious money spread well past the household names, and one US state started treating your brainwaves as sensitive data. Not a quiet week.

The field is getting funded and capable faster than it is getting governed.

1. Meta made non-invasive brain-to-text look serious.

Meta released Brain2Qwerty v2, a system that reads a person's brain activity from a non-invasive scan and turns it into typed text at 61% average word accuracy, up to 78% for the best participant. The peer-reviewed first version landed in Nature Neuroscience on June 29. Say the caveat plainly, because it matters. Participants were physically typing inside a room-sized magnetoencephalography scanner, so the model reads the brain activity of the typing act, not free-floating private thought. It is a research milestone, not mind reading, and not a home device. But non-invasive decoding just moved closer to accuracy people assumed only surgery could reach.

2. The money spread, and the new names matter as much as the dollars.

2026 is a step-change year for BCI capital, roughly $1.7B across about eight deals in the trailing year, on track toward $2B. The biggest rounds went to newer entrants, not the incumbents. Science Corp, led by former Neuralink president Max Hodak, closed $230M at a $1.5B valuation. China's BrainCo raised about $286M, the largest BCI round outside the US, and it is a non-implant wearable player. Alibaba led and Tencent joined a $72.8M round in Shanghai's StairMed. And a startup called Subsense is building a no-surgery interface from nanoparticles delivered by nasal spray. The fixed roster no longer defines the frontier.

3. Connecticut started treating your brain like a medical record.

As of July 1, Connecticut Public Act 25-113 folds neural data into the sensitive-data tier of the state privacy law. Any company reading a resident's brain signals now needs opt-in consent to collect it and separate consent to sell it, with no volume threshold, so a single-user startup is fully in scope. Connecticut is the fourth US state to protect brain data, after California, Colorado and Montana. Notice the pattern. The consumer headbands are being governed before the surgical implants are.

The US had the widest week. The freshest science is Meta's Brain2Qwerty, a non-invasive research release. The freshest concrete clinical event is Paradromics, which completed its first Connexus implant in a patient with motor neuron disease at the University of Michigan on June 17, putting a fourth US invasive program into live human trials, with screening also under way at UC Davis and Mass General. The field also visibly widened past movement. BrainGate reported thought-typing at 22 words per minute with a 1.6% error rate, and Motif Neurotech won an FDA clearance to trial a blueberry-sized skull implant for treatment-resistant depression. Movement, speech and mood are all live BCI targets now. Among the incumbents it was continuity. Neuralink advanced its dura-sparing transdural surgery and continued international PRIME trials, Synchron pushed toward its 2026 pivotal on a $200M Series D, and Precision Neuroscience's surface array has now been used in 85+ patients. The honest caveat has not changed. No paralysis implant is available by prescription in the US in 2026, with realistic timelines running 2028 to 2030.

China's week was about capital and scale more than a single device. StairMed's Alibaba-led, Tencent-joined round backs its 256-channel wireless invasive system, and BrainCo's roughly $286M raise was the largest BCI round outside the US, a reminder the Chinese stack is not only invasive. The structural lead still stands. Neuracle's NEO remains the world's first and only commercially approved invasive BCI, cleared by the NMPA in March, an eight-sensor device that sits on the dura and was built with Tsinghua. NeuCyber's Beinao-1 is now implanted in seven people, and NeuroXess and StairMed continue their human programs. On the clinical-scaling side, hospitals in Tianjin reported expanding use of non-invasive EEG motor-imagery therapy for stroke rehabilitation. BCI sits in China's latest five-year plan as one of six priority future industries.

Europe builds real hardware and it showed this week, so do not file it under regulation only. Germany's CorTec reported a stroke-patient result with its implantable Brain Interchange system. Spain's INBRAIN reported first-in-human results from a graphene-based interface used during tumour-resection surgery, and completed enrollment of that world-first study. Switzerland's Neurosoft Bioelectronics closed a $7.5M seed. The financing gap is the real weakness, not the science. North America holds the overwhelming share of disclosed neurotech capital while Europe's recent disclosed rounds are small. Neurotechnology is now the single largest slice of the European Innovation Council's portfolio, near 14%, which is where the next European hardware bets are being seeded. The governance layer remains the durable European signal. The EU AI Act treats real-time neural-data interpretation and BCIs used as safety components as high-risk, and the case for a dedicated EU neurotechnology strategy is building.

The new name this week is Fluent, a University of Melbourne spin-off building a minimally invasive speech interface called InScribe that reads the neural signals of attempted speech and turns them into text. It sits under the scalp but outside the skull, aimed at people who lose their voice to ALS or MS. It is early-stage, a company profile rather than a clinical result, but it is real and worth watching. It also fits a pattern. With Synchron's Stentrode originating at Melbourne, Coherence Neuro's SOMA-1 cancer-sensing implant tested at Royal Melbourne Hospital in June, and now Fluent, Australia keeps producing BCI companies. Across the wider Asia-Pacific the structural read holds, the fastest-growing BCI region, driven by national programs, semiconductor supply chains and streamlined approval pathways.

Two threads defined the week. The first is non-invasive decoding getting more capable. Meta's Brain2Qwerty v2 reached 61% average word accuracy from magnetoencephalography while participants typed, trained on about 22,000 sentences from nine volunteers, with the peer-reviewed v1 in Nature Neuroscience on June 29. Yale's manifold-geometry result, published earlier in June, showed people learn to control a non-invasive BCI far faster when the decoder follows the brain's own natural geometry, reframing training as fitting the machine to the brain rather than the reverse. The second thread is durability. The Nature Medicine at-home ALS case remains the clinical reference point. One participant used an intracortical implant independently at home for more than 3,800 hours across nearly two years, communicating more than 180,000 sentences while holding a full-time job. Together they mark a shift, decoders that are both more humane and more sample-efficient, and implants that work for years rather than for a demo.

The live change is Connecticut Public Act 25-113, in force since July 1. Neural data added to the sensitive-data category, consent to process, separate consent to sell, no volume threshold. Connecticut is the fourth US state to draw this line. The tension worth watching is that the law protects measured brain data while the treatment of inferences drawn from those signals is still unsettled. There is still no federal neural-data statute. The MIND Act, introduced late in 2025 to direct the FTC to study neural-data privacy, has not advanced beyond committee, so Section 5 of the FTC Act remains the only federal backstop. That gap is sharper now that roughly $2B of 2026 capital is flowing into devices designed to decode mental content, including non-surgical approaches that could reach consumers faster than implants. Chile's constitutional neurorights amendment remains the global benchmark.