Paradromics's first Connexus participant used a brain implant to speak her own words

By Inside BCI · September 16, 2026

“Okay, I have a lot to say.” Those seven words are the first free sentence a Michigan woman in her 60s used to speak in real time through Paradromics’s Connexus intracortical brain-computer interface, three months after her June 2026 implantation at University of Michigan Health. Paradromics announced the milestone on 14 September 2026. The participant has progressive motor neuron disease and severe dysarthria, retains limited natural speech, and is the first participant in the company’s Connect-One Early Feasibility Study.

The decoder output is text plus synthesized speech through an external device. It is not avatar-mediated. Both disclosed verbatim sentences (the second: “Don’t be afraid. It’s a wonderful experience to be able to help others by your participation in the research study.”) were decoded word for word, in real time, with no errors. The participant has since answered open-ended questions and held a live phone call with her grandchildren.

What was disclosed and what was not

The disclosure is a demonstration, not a benchmark. Paradromics did not release words-per-minute, word-error rate, vocabulary size, or decoding latency figures. The Connect-One protocol targets 6-year follow-up per participant. The three-site study also runs at UC Davis Health under Lead Principal Investigator David M. Brandman, MD, PhD, and at Mass General Brigham Neuroscience Institute under Daniel Rubin, MD, PhD. The FDA Investigational Device Exemption for Connect-One was approved in November 2025.

The absence of quantitative benchmarks matters. Prior BCI speech-decoder results in the peer-reviewed literature (Willett et al. BrainGate 2023: approximately 62 words per minute at 23.8 percent word-error rate on 125,000-word vocabulary; UCSF Chang team 2023 avatar work: approximately 78 words per minute) set the reference frame for what real-time speech-BCI performance looks like at published benchmark scale. Paradromics’s 14 September disclosure does not enter that comparison in numeric terms. Two clean sentences per released demonstration is a proof of the pathway from cortex to spoken word, not a claim of decoder throughput.

The implant and the operating room

Paradromics’s Connexus device is a fully implanted intracortical BCI comprising 421 platinum-iridium microwires arranged in a dime-sized cortical array. A lead runs under the skin from the array to a sub-clavicular power and data unit, which communicates wirelessly through intact skin to an external receiver.

The University of Michigan surgery, performed by Matthew Willsey, MD, PhD, took approximately four hours. The approach was a craniotomy through an inverted U-shaped incision. The array was placed in the precentral gyrus, in the region of motor cortex that controls the muscles used to pronounce speech sounds. Post-surgical testing moved to the participant’s home, where the family phone call and open-ended question-answering sessions have taken place.

The three-month arc from implant to speech

Paradromics’s Connect-One EFS first-in-human implantation was 17 June 2026. On 26 August 2026, the FDA approved expansion of the Connect-One protocol to allow the Connexus system to interface with the participant’s own personal computing devices through Paradromics’s Convey software layer, as InsideBCI covered at the time. The 14 September real-time speech milestone is a distinct achievement on a separate technical track: the 26 August approval concerned the regulatory pathway for endpoint device compatibility, while the 14 September result reports decoder performance sufficient to produce the participant’s chosen words in real time.

The full arc from implantation to first free sentence has taken approximately three months.

What Angle and Brandman said

Matt Angle, PhD, founder and CEO of Paradromics: “When we implanted Connexus in June, we crossed the threshold into long-term human use. We have now crossed another: our first participant is choosing her own words and using neural activity to communicate them in real time. She determines what she wants to say, and the system enables her to say it.” Angle separately positions the Paradromics platform as “effectively the world’s most sophisticated conduit between human intention and AI.”

Angle to CBS Austin: “The same kinds of devices that can read out intended speech can read out intended movement… almost untreatable conditions… become solvable technology problems.” The technical read is that a cortical decoder trained on speech-related motor-cortex activity is in principle extensible to hand and arm motor control from adjacent motor-cortex territory, though Paradromics has not yet announced a motor-control clinical milestone.

David Brandman, MD, PhD, Lead Principal Investigator (UC Davis Health): “The very first participant implanted with the Connexus system has been empowered to say whatever she wants to say using her brain-computer interface. This is a landmark milestone for the field.”

Chief Clinical Officer William J. Marks Jr., MD, MS-HCM, who joined Paradromics on 7 July 2026 with a background covering Verily Head of Clinical Neurology and Medtronic global DBS medical direction, is not quoted in the 14 September release. Chief Technology Officer Kurtis Nishimura is also not quoted.

Same-day academic milestone: UCSF Chang team, Nature Neuroscience

On the same day, Edward F. Chang’s UCSF group published in Nature Neuroscience (DOI 10.1038/s41593-026-02446-2, “Simultaneous speech and gesture decoding for multimodal communication in paralysis”) a demonstration that a single cortical implant can decode speech and gesture simultaneously to animate a virtual avatar. NIH-funded. The Chang result is academic, uses avatar-mediated output, and was published in a peer-reviewed journal.

The Paradromics result is commercial, uses text plus synthesized speech, and was released as a company announcement backed by a Business Wire press release. The two announcements landed within 24 hours and share the “restore communication in paralysis” indication, but they are categorically different achievements from different institutions using different output modalities and different verification standards. They should not be conflated. That both landed in the same 24-hour cycle is coincidental, and useful only insofar as it signals the pace at which speech-BCI results are now surfacing in September 2026.

What Neuralink and Synchron have not been said to do

The 14 September disclosure does not compare Connexus performance to Neuralink’s Prime study (which centres on cursor and text output via the N1 implant) or to Synchron’s COMMAND / SWITCH programme (which centres on communication and control via the endovascular Stentrode). Paradromics is running its own line. A cross-operator comparison would require Neuralink and Synchron to have published equivalent participant-scale performance data on comparable indications, which they have not.

Not disclosed

Paradromics did not release quantitative decoder metrics (words per minute, word-error rate, vocabulary size, latency from thought to output). The participant’s specific motor neuron disease diagnosis (ALS versus PLS versus another MND subtype) was not disclosed; the release names ALS and PLS only as umbrella examples of the class. The participant’s identity was not disclosed. Post-milestone enrollment schedule for participants 2 and 3 across the three Connect-One sites was not disclosed. Timeline to pivotal trial and IDE-to-PMA regulatory transition was not disclosed. No commercial pricing or reimbursement pathway signal was provided.

Tier-1 English media coverage as of publication was thin: The New York Times ran a “Well” section feature, CBS Austin ran a local Austin-company piece, and Business Wire syndication carried the announcement to Yahoo Finance Singapore, Eagle-Tribune, and Anadolu Ajansı. Bloomberg, Reuters, WSJ, FT, STAT, TechCrunch, MIT Technology Review, FierceBiotech, and Endpoints News did not appear to run the 14 September story as of publication. This is a quieter reception than Paradromics’s November 2025 IDE approval received.

Where this sits in the invasive-BCI operator landscape

Paradromics is the third named US commercial invasive-BCI operator (alongside Neuralink and Synchron, with Blackrock Neurotech’s Utah array as the long-running Class III precursor and Precision Neuroscience’s Layer 7 cortical array as the epicortical peer) to demonstrate a real-time speech decoding result in a first-in-human study participant. The result confirms that the Connexus 421-microelectrode intracortical BCI can produce useful cortical signal from the speech-motor region of the precentral gyrus at a signal quality sufficient for real-time speech decoding. It does not yet confirm decoder performance at benchmark scale, which will require the peer-reviewed follow-on publication that the field will now expect.

Read alongside the same-week Nature Neuroscience multimodal result from UCSF (Chang team) and the 8 September 2026 Nature Neuroscience Comment on implantable BCI research ethics (Miller-Schalk et al., which InsideBCI covered), the September 2026 news cycle has produced three distinct speech-BCI signals in nine days: a commercial demonstration (Paradromics), an academic peer-reviewed multimodal breakthrough (Chang), and a research-ethics call for tighter discipline as implant numbers scale (Miller-Schalk). The commercial-development curve, the academic-benchmark curve, and the ethics-frame curve are all accelerating in parallel.

What to watch

Whether Paradromics publishes peer-reviewed decoder performance metrics (words per minute, word-error rate, vocabulary size) from the Michigan participant in the coming quarters. The 14 September release is a demonstration; peer-reviewed benchmark data is the next hard signal.

Whether the second and third Connect-One participants are enrolled at UC Davis Health (Brandman) or Mass General Brigham (Rubin) in Q4 2026 or H1 2027. Multi-site multi-participant enrollment is what moves an EFS from proof of concept to a pivotal-trial-eligible dataset.

Whether Paradromics announces a motor-control (arm and hand) clinical milestone from the same participant’s Connexus implant. Angle’s CBS Austin remarks specifically framed the platform as extensible from speech to motor control. A next-quarter motor-control demonstration would confirm the 421-electrode intracortical array is decoder-general across adjacent motor-cortex territories.

Whether Neuralink or Synchron publish comparable participant-scale speech-decoding results in the next two quarters. Paradromics’s 14 September milestone is a category signal for the US commercial invasive-BCI operator group. A peer response is now on the clock.

Whether the Miller-Schalk Nature Neuroscience Comment’s four ethical priorities (research versus care distinction, long-term participant support, participant risk-benefit fairness, meaningful clinical purpose) are picked up as reference language in the Paradromics Connect-One study protocol or in the iBCI-CC’s next public output.

Sources

Paradromics 14 September 2026 milestone:

Prior Paradromics coverage referenced:

Same-week academic milestone:

Cross-reference to prior InsideBCI coverage:

news Paradromics Connexus Connect-One Matt-Angle Vikash-Gilja Matthew-Willsey David-Brandman Daniel-Rubin University-of-Michigan-Health UC-Davis-Health Mass-General-Brigham motor-neuron-disease MND ALS PLS dysarthria real-time-speech intracortical-BCI FDA-IDE EFS precentral-gyrus 421-microelectrodes Edward-Chang-UCSF Nature-Neuroscience Issue-01