Imagine being able to communicate without speaking out loud. That’s what researchers at Stanford are working on with brain-computer interfaces, or BCIs. These devices can pick up signals from your brain and turn them into words or commands. The goal is to help people who can’t speak due to paralysis. But there’s a catch—what if your private thoughts leak out?

Most BCIs designed for speech work by decoding attempted speech. That’s when people try to speak but can’t move their muscles. The device picks up signals from the brain areas that control muscles and translates them into words. This method works well for people who can still try to speak, but it’s hard for those with severe paralysis. So Stanford’s team decided to go a step further and focus on inner speech—that quiet, mental chatter we all have.

Decoding Inner Speech in the Brain

The researchers collected brain data by implanting tiny electrode arrays in the motor cortex of four paralyzed participants. These volunteers listened to recordings or read silently. The team looked for patterns in the brain signals that corresponded to inner speech. Interestingly, they found that the same brain regions involved in attempted speech also showed activity during inner speech. That meant the system trained to decode attempted speech could sometimes accidentally pick up inner thoughts.

This raised an important concern: could a BCI unintentionally reveal private thoughts? To address this, the team developed safety measures. One approach involved training AI algorithms to distinguish between attempted speech and inner speech signals. By labeling inner speech as “silent,” the AI could learn to ignore these signals. This way, the system only decoded what the user intended to say aloud, not what they thought silently.

Protecting Mental Privacy with Safeguards

The researchers also implemented a special “mental password” system. Users had to imagine saying a specific phrase—like “Chitty chitty bang bang”—to activate the prosthesis. The system recognized this phrase with 98% accuracy. This acts as a kind of lock, ensuring only the right person can control the device. Complex sentences or unstructured thoughts, however, proved much harder to decode accurately.

Testing with simple, cued words showed promising results. Patients could imagine saying short sentences, and the system correctly identified them around 86% of the time with a small vocabulary. But as the vocabulary expanded, accuracy dropped. When they tried to decode more natural, unstructured inner speech—like thinking about a favorite food or a quote—the system struggled. The outputs were often gibberish, and success rates hovered just above chance.

Despite these challenges, the work is a big step forward. It shows that decoding inner speech is possible, at least in controlled conditions. But the high error rates mean it’s not ready for everyday use yet. Krasa, the lead researcher, points out that hardware limitations play a big role. More electrodes or better signal quality could improve accuracy. Also, inner speech might be more clearly represented in other brain regions outside the motor cortex.

Future Directions and Privacy Concerns

The team is now exploring how much faster an inner speech BCI could be compared to traditional attempted speech systems. They’re also interested in helping people with aphasia—a condition where they can control their mouths but can’t produce words. The idea is that decoding inner speech might restore some form of communication for them.

The research raises important questions about mental privacy. If BCIs become more accurate, what safeguards will protect our thoughts? Right now, the team’s methods—such as training AI to recognize passwords—are simple, but future technology could be more invasive. It’s essential to develop strong protections so that our inner worlds remain private.

In sum, decoding inner speech shows huge potential. It could someday give people who can’t speak a new way to communicate. But it also reminds us to think carefully about how to keep our mental privacy safe as this technology develops. For now, it’s a promising proof of concept, with many hurdles to overcome before it becomes part of everyday life.

Inspired by

• https://arstechnica.com/science/2025/08/an-inner-speech-decoder-reveals-some-mental-privacy-issues/

Sources

• conde.digital
• cnevids.com
• doi.org