Brain-computer interfaces today: Real communication breakthroughs, stubborn engineering limits, and the hype traps to avoid

What BCIs can do now

Brain-computer interfaces (BCIs) have crossed a threshold that matters. In carefully controlled research settings, people who cannot speak or move normally have used implanted systems to generate text at speeds that begin to resemble everyday communication, and to control computers with more independence. A UK Parliamentary briefing captures the core idea: BCIs record brain activity and translate it into commands for external devices.

The most striking progress is in communication. In one Nature study, a participant with paralysis produced text by attempting handwriting, reaching 90 characters per minute (cpm). In another, researchers reported decoding unconstrained sentences at 62 words per minute (wpm) using implanted microelectrode arrays. A separate Nature paper describes a brainstem-stroke survivor using a high-density electrocorticography (ECoG) array to decode speech-related signals into text, synthesised voice, and a virtual avatar.

Those results are real, and they matter. They also sit at the far end of a trade-off triangle, performance, invasiveness, and practicality.

What a BCI is, and the three ways it is built

A BCI is best understood as three parts: a sensor that captures brain signals, a decoder that translates those signals into meaning, and an output device, such as a cursor or speech synthesiser.

BCIs are commonly grouped by where the sensors sit.

Non-invasive BCIs place sensors outside the skull, often using electroencephalography (EEG). EEG headsets are comparatively low risk, but the signals are weak and noisy, which limits the information bandwidth.

Semi-invasive BCIs sit closer to the brain without penetrating brain tissue, or use alternative routes. A notable example is an endovascular approach, recording signals from within a blood vessel next to cortex. A JAMA Neurology study reported safety and feasibility for this approach in a first-in-human setting.

Invasive BCIs involve surgical implantation of electrodes under the skull, including arrays that penetrate cortex. These can provide high-resolution signals that support faster and more precise decoding, but raise surgical, infection, and long-term device considerations.

The signals: why performance comes with trade-offs

The difference between a viral headline and a usable daily tool often comes down to signal quality and stability.

EEG measures electrical activity through skull and scalp. It is accessible, but noisy. ECoG places electrodes on the brain surface, improving signal fidelity. Intracortical arrays can record at or near single-neuron resolution, enabling strong performance in some tasks, but with greater invasiveness.

Even with strong signals, decoding is not simply “reading thoughts”. Most high-performing systems decode intended movement, such as handwriting motions, or the motor planning involved in speech. That distinction matters because it explains both the promise and the limits. It can restore communication channels, but it does not deliver general mind reading.

Communication: the headline use case, with caveats

Communication BCIs now split into two broad strategies.

Brain-to-text systems translate neural activity into letters and words. The handwriting approach is notable because it uses a complex movement pattern. Researchers argue those patterns can be easier to decode than point-and-click cursor movements, contributing to higher typing rates.

Speech decoding systems aim for more natural conversation. Intracortical arrays have enabled decoding of unconstrained sentences at 62 wpm. ECoG-based systems have paired speech decoding with avatar control, targeting not only words but expression. Scientific Reports has also described online speech synthesis using chronically implanted ECoG in a man with ALS, a reminder that “online” operation is a major step toward real use.

These systems remain experimental, typically tested with one person or a small cohort, with substantial training and specialist support. They demonstrate what is possible, not what is routine.

Cursor control and rehabilitation: the quieter successes

Cursor control is a foundational capability. BrainGate research demonstrated neural control of cursor trajectory and clicking using an intracortical array in people with tetraplegia. This work matters because it translates into practical computer access, selecting letters, navigating interfaces, controlling assistive devices.

Rehabilitation applications often use BCIs as part of therapy rather than as a permanent replacement for movement. Reviews in clinical and rehabilitation contexts describe BCIs as tools that may restore or supplement function, but also stress challenges around robustness and translation from lab to clinic.

What remains hard

Three barriers keep reappearing.

Reliability over time: Signals drift, and decoders must cope with day-to-day variation. A system that works brilliantly on day one can degrade, or require frequent recalibration.

Training burden: High performance typically requires user training and data collection. That is not a minor inconvenience, it is often the dominant cost in time and support.

Latency and interaction quality: Communication feels natural only when delay is low and errors are manageable. Research is pushing toward lower-latency, streaming approaches, but robustness is still a hurdle.

Risks: surgery, infection, privacy, and long-term support

Invasive systems involve surgery and the associated risks. Semi-invasive approaches like endovascular recording aim to reduce surgical burden, but they are not risk-free and still require careful clinical oversight.

Security is an underappreciated risk. BCIs combine implanted or wearable sensors with external computers and software. The FDA’s cybersecurity guidance for medical devices underscores that connected devices must be designed and reviewed with cyber resilience in mind.

Privacy and consent risks expand as neural data becomes a commercial asset. The UK POST briefing flags privacy and responsibility as core ethical questions. UNESCO’s new ethical framework on neurotechnology adds weight to concerns about neural data and human rights safeguards.

Long-term support is where hype often collapses. Devices need maintenance, upgrades, and sometimes removal. A 2025 review highlights ethical and practical considerations around explantation, a reminder that responsibility does not end at implantation.

What would count as a breakthrough

A genuine breakthrough would not be a faster demo in one person. It would be a step change in dependable daily use, across many users, with manageable training and clear pathways for long-term support.

In practice, that means: stable performance for months, easy recalibration, safe home use, lower-burden implantation or high-performing non-invasive sensing, and governance that treats neural data as highly sensitive.

Hype traps to avoid

The safest rule is to ask what the demo does not show: how often it fails, how much retraining it needs, what happens at home, and who pays for support after the research grant ends. BCIs are advancing, but the hard part is turning extraordinary prototypes into ordinary reliability.

Fact-check list: claims, sources, confidence

• BCIs record brain activity and translate it into commands for external devices. High
• BCIs are commonly categorised as invasive and non-invasive, with EEG used in non-invasive headsets. High
• Endovascular BCI feasibility and a favourable safety profile were reported in the first-in-human SWITCH study. High
• Brain-to-text via attempted handwriting reached 90 cpm in a participant with paralysis. High
• A speech BCI decoded unconstrained sentences at 62 wpm using microelectrode arrays. High
• A high-density ECoG array enabled decoding to text and avatar control in a brainstem-stroke survivor with anarthria. High
• Online synthesis of intelligible words using chronically implanted ECoG in a man with ALS has been reported. High
• Intracortical arrays have enabled cursor trajectory and click control in people with tetraplegia. High
• The FDA has final guidance on cybersecurity for medical devices, including design and documentation recommendations for devices with cybersecurity risk. High
• The UK POST briefing highlights privacy and responsibility as ethical challenges for BCIs. High
• UNESCO adopted a global ethical framework on neurotechnology, including safeguards around human rights and consent. High
• Explantation and end-of-life responsibilities are emerging ethical issues for neural devices. Medium (scope varies by device type and jurisdiction)
• Neural data protection may be conceptualised within privacy frameworks such as GDPR, but interpretation is evolving. Medium (jurisdictional and legal interpretation varies)