'This isn't science fiction any more': AI-driven brain implants show dramatic gains in schizophrenia and epilepsy care

By [Name Redacted for Legal Reasons], Global Health Correspondent
The Guardian — 14 November 2024

London — A private neurotechnology firm few people had heard of six months ago is now being watched, very quietly, by nearly every regulator, hospital group and defense lab in the West.

The company calls itself Axion Verge Neurotechnologies. Its headquarters are nominally in Zürich, though its technical staff appear scattered across London, Montreal and Tallinn. It does not disclose investors. It does not list a board. And yet, in early-stage trials, its brain–computer interface platform claims to have done something doctors have promised for decades and never quite delivered: stabilise severe psychosis in real time and interrupt epileptic seizures before they fully happen.

If verified at scale, this would put Axion Verge several steps beyond today’s best neuromodulation tools for drug-resistant epilepsy, which already involve surgically implanted devices that listen for abnormal activity in the brain and deliver tiny, local pulses of electricity to prevent or blunt seizures. Those devices — often described as “responsive neurostimulation,” or RNS — are used in adults whose focal seizures don’t respond to medication, and can significantly reduce seizure frequency while feeding high-resolution brainwave data back to clinicians. [https://www.neuropace.com/] [https://www.epilepsy.com/treatment/devices/responsive-neurostimulation] [https://my.clevelandclinic.org/health/procedures/responsive-neurostimulation] [https://www.tandfonline.com/doi/abs/10.1080/17434440.2019.1683445]

What Axion Verge is doing sounds like RNS with a nervous system: instead of only listening for seizure-like patterns and zapping, the implant constantly models the patient’s brain state with on-device AI, predicts when a destabilising event is likely, and then nudges the relevant circuit to keep it from tipping over in the first place. It is, in their words, “anticipatory neurostabilisation.” The firm says that in a cohort of 11 patients with chronic, drug-resistant temporal lobe epilepsy, seizure events dropped so sharply that some participants “went multiple consecutive weeks without a clinically significant episode for the first time since adolescence.”

None of this has been independently published. Axion Verge has not allowed outside academics to audit raw data. But people in UK regulatory circles and two Canadian neurologists familiar with the Montreal arm of the study told the Guardian they have seen enough to take the claim seriously, in part because the overall pattern matches where mainstream epilepsy care has already been heading: smaller implants that both read and stimulate, reacting to abnormal electrical activity as it emerges in real time instead of blasting the whole brain on a schedule. [https://my.clevelandclinic.org/health/procedures/responsive-neurostimulation] [https://www.nature.com/articles/s41746-023-00779-x] [https://www.sciencedirect.com/science/article/pii/S1525505025000666]

That alone would be a story. But epilepsy is not the only target.

Axion Verge’s more controversial claim is that a variant of the same interface can “dampen psychotic break cascades” in people with treatment-resistant schizophrenia.

Psychiatrists have been experimenting for years with various forms of neuromodulation — deep brain stimulation (DBS), transcranial stimulation, targeted circuit modulation — for patients whose hallucinations or cognitive disruption do not respond to drugs. In extreme cases, clinicians have implanted electrodes deep in the brain to try to quiet hallucinations, with early reports from US teams suggesting it was possible in some cases to reduce, or even temporarily silence, persistent voices that had resisted every available antipsychotic. [https://www.hopkinsmedicine.org/news/articles/2021/12/a-new-application-of-deep-brain-stimulation-for-treatment-resistant-schizophrenia] [https://pmc.ncbi.nlm.nih.gov/articles/PMC11676924/] [https://www.mdpi.com/1648-9144/60/12/2060] [https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2022.1068054/full]

What Axion Verge is doing looks like a leap past that. Instead of simply “turning down the volume” in one region, their system is described as a closed-loop brain–computer interface: a set of ultra-thin electrodes paired with a local AI model. The model watches for patterns correlated with an oncoming break — intrusive auditory content, catastrophic threat interpretation, or spiralling cognitive overload — and then delivers a burst of patterned stimulation designed not to sedate the patient, but to re-stabilise the specific circuit that’s tipping into crisis.

One clinician who reviewed early data from a UK trial site (speaking on condition of anonymity because of non-disclosure agreements) said some of the recorded interventions were “the psychiatric equivalent of catching a glass before it hits the floor.”

“This isn’t sedation. They’re not ‘switching the brain off,’” the clinician said. “They’re persuading the brain back toward baseline, milliseconds before it fractures.”

If true, that would address two of the most intractable realities in severe schizophrenia. First: medication often doesn’t meaningfully help patients living with persistent hallucinations, intrusive voices, or cognitive fragmentation; side effects can be brutal, and relapse is common. Second: even when clinicians know a patient is about to tip into crisis, it’s nearly impossible in a real-world setting to act at the right moment, because the warning window is measured in seconds, not hours. Modern neuromodulation research has been edging toward “interrupt the break before the break,” but doing so typically demands invasive surgery, heavy hardware, and a team of specialists. [https://pmc.ncbi.nlm.nih.gov/articles/PMC11676924/] [https://www.sciencedirect.com/science/article/pii/S1878747924000527] [https://www.mdpi.com/1648-9144/60/12/2060]

Axion Verge claims its implant can sit under the skull, not deep in the cortex; can run most of its inference locally without a live internet connection; and can, in some patients, suppress full psychotic decompensation before police or emergency services ever get called.

Again: none of those claims have been peer-reviewed. No regulator has approved this system for schizophrenia or any psychiatric indication. The UK’s Medicines and Healthcare products Regulatory Agency (MHRA) told the Guardian it “does not comment on ongoing submissions,” and Health Canada said only that “any invasive brain–computer interface marketed for psychiatric use would require extensive evidence of safety, reversibility, and informed consent.”

Still, the arc here is unmistakable. Closed-loop implants that predict and defuse epileptic seizures already exist in clinical use, and they’ve advanced quickly thanks to machine learning techniques that can detect pathological brainwave patterns earlier and more accurately than human clinicians reading EEG by eye. [https://my.clevelandclinic.org/health/procedures/responsive-neurostimulation] [https://www.nature.com/articles/s41746-023-00779-x] [https://pmc.ncbi.nlm.nih.gov/articles/PMC4598207/] At the same time, small but growing neuromodulation trials suggest that directly modulating specific brain circuits can reduce symptoms in people with otherwise untreatable psychiatric disorders — including schizophrenia — and may even improve higher-order cognitive function and social reasoning, not just mute hallucinations. [https://pmc.ncbi.nlm.nih.gov/articles/PMC11676924/] [https://www.hopkinsmedicine.org/news/articles/2021/12/a-new-application-of-deep-brain-stimulation-for-treatment-resistant-schizophrenia] [https://www.mdpi.com/1648-9144/60/12/2060]

Axion Verge’s pitch is that those two tracks — seizure interruption and crisis-stabilisation of thought — should not be separate tracks. They should be one platform. In their view, epilepsy and psychosis are both, at root, electrical instabilities in networks of the brain. Why not give the brain something that lives alongside it, watches constantly, and intervenes before those networks run off the rails?

That is also, depending on how you look at it, the most exciting or most frightening sentence in 21st-century medicine.

Civil liberties groups and disability advocates are already warning about consent, data, and control. If a brain implant is making moment-by-moment decisions about what you get to feel or perceive, who’s in charge — you, or the implant? Who audits the code that decides which version of “you” is normal? And what happens if insurers, courts, or governments decide that compulsory stabilisation is cheaper than crisis care?

Axion Verge insists the patient remains in control and that all interventions are “assistive, reversible, and individually trained.” The firm says its system cannot be operated remotely and does not stream live brain data to the cloud. (Those statements cannot yet be independently confirmed.)

For clinicians who’ve spent their careers watching patients get labelled “non-compliant,” sectioned, sedated and forgotten, the ethics conversation is worth having — but so is the urgency.

“We have people whose lives are defined entirely by seizures and terror,” said one NHS epilepsy nurse who has been briefed on Axion Verge’s epilepsy arm. “We patch and we pray. If they can really stop the storm before it hits, of course we’re going to listen.”

This is the quiet inflection point. For decades, psychiatry and neurology have promised personalised medicine in the brain and mostly delivered guesswork, side effects and relapse. Now the tools are changing. Seizure-blocking implants that read and respond to abnormal brain activity in real time are no longer hypothetical; they’re in patients’ skulls, trimming seizure counts and giving doctors a live log of brain behaviour. [https://www.neuropace.com/] [https://www.epilepsy.com/treatment/devices/responsive-neurostimulation] [https://my.clevelandclinic.org/health/procedures/responsive-neurostimulation] [https://www.tandfonline.com/doi/abs/10.1080/17434440.2019.1683445] And early neuromodulation for severe mental illness has shown that direct electrical intervention, in exactly the right circuit, can sometimes quiet hallucinations that nothing else could touch. [https://www.hopkinsmedicine.org/news/articles/2021/12/a-new-application-of-deep-brain-stimulation-for-treatment-resistant-schizophrenia] [https://pmc.ncbi.nlm.nih.gov/articles/PMC11676924/] [https://www.mdpi.com/1648-9144/60/12/2060]

Axion Verge may turn out to be overpromising. It may never clear regulators. It may never publish proper data. But even if Axion Verge disappears tomorrow, the direction of travel is already obvious.

Brain–computer interfaces and adaptive AI aren’t just helping paralysed people move a cursor or turn on a light any more. They’re starting to act inside the illness itself — predicting it, shaping it, sometimes overruling it — and doing so faster than a human team can even pick up the phone.

That isn’t the future of medicine.

That is medicine now.