Click the link below the picture
.
Half an hour before bed, Adnan Jinnah sits on a chair and straps on a headband. It looks like something a tennis player might wear, save for a large rectangular button on the front. When he feels ready to sleep, he lies down, presses the button, and starts an app on his phone. “That’s pretty much it,” he says.
As he sleeps, sensors in the device pick up Jinnah’s brain activity, heart rate, and blood flow and oxygen to his brain. The signals are then interpreted by algorithms that spit out data on his sleep quality that he can look at in the morning.
The device, known as a Muse headband, was developed by the Canadian consumer neurotechnology company Interaxon. On and off for more than 4 years Jinnah, who lives in Leicester, England, has had trouble falling asleep. He says daily reports from the device allow him to monitor how well he slept the night before and how poor sleep affects his mood. It also provides him with long-term data about his sleep that have helped him understand how different factors, including medication or stress, can influence his rest.
What Jinnah didn’t know, however, was that data from his headband may have been feeding an artificial intelligence (AI) model of the brain. Interaxon is building the model—known in the industry as a “brain foundation model”—to sharpen its picture of how the brain works, especially during sleep and times of focus. The company says this would allow it to provide a better service to its users and develop a product to help support brain research. The data used to train it are overwhelmingly coming from users like Jinnah.
So far, Interaxon has trained its model on about 50,000 hours of electroencephalography (EEG) data from users of the technology, the company says. The data are stripped of personal details, such as people’s names and emails. The company plans to keep expanding this figure. “We aim to use as much of that eligible corpus as possible,” says Chief Marketing Officer Nadia Kumentas.
Yet Interaxon’s privacy policy did not explicitly tell users their data might be used to train AI brain models. After Science pointed this out, the company updated the policy. “We agree users should be explicitly informed,” Kumentas says.
For Jinnah and potentially tens of thousands of other Muse users like him, it was too little too late: The training had already happened, without his explicit informed consent. When user data get embedded into an AI system they are near-impossible to remove.
Jinnah felt a sense of “resignation” when Science told him his neural data may have been used to train an AI system that he had never heard about. “I suppose it’s very similar to when big tech companies consume your data.” But he thought users should be asked for their consent.
Interaxon is not unusual. In a yearlong investigation, Science reviewed dozens of consumer neurotech privacy policies and spoke to neurotech experts, companies, and users to understand how consumers’ brain data are being used around the world. The investigation reveals a largely unregulated global ecosystem in which tens of millions of hours of EEG data have already been collected by consumer neurotech companies. Multiple firms are using these data in the race to build a brain foundation model.
The findings come as researchers and policymakers argue that the massive accumulation of brain data constitutes a broader privacy risk. Multilateral organizations such as UNESCO and the Organisation for Economic Co-operation and Development have recently highlighted the rise of consumer-grade brain-computer interfaces as a special concern. In October, United Nations Secretary-General António Guterres presented a report warning that the rapid development of neurotechnologies could give governments access to our innermost thoughts, posing a threat to human rights.
Some experts argue that because consumer neurotechnology devices don’t collect clinical-grade data, it’s unlikely that companies—or governments with access to the data—can glean much about users. But most agree that users have the right to know how their neural data are being used, and that more needs to be done to protect the data.
“Brain signals aren’t heartbeats. Monitoring them is meaningfully different,” says Owen Jones, professor of law and biological sciences at Vanderbilt University. “We are the sum of our brain’s activities in a way that we are never the sum of our other organs’ functions.”
Over the past decade, wearable neurotechnologies have caught on around the world. Consumers can buy devices that promise to monitor stress, fatigue, focus, sleep, meditative states, and more from a growing number of companies. The neurotechnology market value is also steadily on the rise: It is estimated at about $19.19 billion this year and expected to reach nearly $33 billion by 2034. Recently, tech giants have also entered the arena. In 2023, Apple filed a patent for earbuds that could collect EEG data. And last year, Meta launched its Neural Band, which collects data on tiny electrical signals from wrist muscles.
The majority of these wearable neurotechnologies are stripped-down versions of EEG headsets commonly used in hospitals. Clinical-grade headsets tend to gather data using 19 to 25 electrodes spaced across the skull, but brain wearables usually have between one and six. These generally cover a limited part of a person’s brain region—for example, the front of the head or around the ears—depending on what the device is meant to track.
Although some devices have been rigorously tested and used in thousands of scientific studies, the vast majority lack scientific underpinning. One study published in 2018 examined the scientific evidence behind claims made by consumer EEG companies that providing people with real-time information about their brain waves can help them improve their concentration, sleep, and other behaviors. Of more than 3000 publications cited by these companies, only 11 used gold-standard research methods, and 10 of these found no real benefits compared with a control condition.
Because consumer-oriented devices tend to be lighter, simpler, and more user-friendly than clinical tools, the quality of the data is often compromised. EEG is also inherently limited. It’s great for understanding when brain activity takes place, but not for telling you exactly where, particularly when the activity originates in deeper parts of the brain, such as the amygdala, hippocampus, or thalamus, which play important roles in emotion, memory, and sensory processing. As such, some experts say that although brain data collected by consumer devices might reveal broad changes in neural activity, such as when someone falls asleep, they can’t reveal much about a person’s mental state.
But although EEG might not read people’s thoughts, “it would be equally misleading to suggest that it cannot reveal meaningful information about their cognitive states,” says Pietro Aricò, professor of bioengineering at the Sapienza University of Rome. And as consumer EEG devices begin to incorporate AI and are able to conduct increasingly sophisticated analyses, data collected on the outer surface of the brain could be enough to decode intimate features of our mental lives, including our thoughts, emotions, and personality, says Rafael Yuste, co-founder of the Neurorights Foundation, a nonprofit dedicated to protecting people’s brain data rights.
For the companies that make these devices, the data are gold, and they have amassed millions of hours of them. They are used to train algorithms, help companies offer personalized insights to users, and generate evidence they can use to market their products. “Our brain data is the most valuable part of the company,” Kumentas says.
Increasingly, brain data are valued for a specific purpose: training AI brain models. In many ways these models are analogous to large language models such as ChatGPT or Claude, but they are trained on neural data rather than text. After being fed an enormous amount of data, a model recognizes key patterns of brain activity. Researchers can then easily adapt this model to perform different tasks, such as identifying patterns associated with sleep or neurological conditions. The goal is to build a flexible model of the brain that can be used for various tasks with little or no additional training.
The companies building these models believe they could become a breakthrough technology, by helping researchers better understand how the brain works and automating some doctors’ tasks so they can focus on more critical work. “The model is there in order to serve the world for so many things,” said Tan Le, founder and CEO of Emotiv, a San Francisco–based neurotech startup that has developed multiple EEG wearables.
These models are still at a relatively early stage. But they have already shown promise. The AI neurotech startup Piramidal, for instance, has developed a brain foundation model using EEG data from publicly available sources and hospital partnerships, rather than consumer devices. Last year, the startup partnered with the Cleveland Clinic to develop a way to automatically detect seizures in hospital patients. After the model was fine-tuned on the clinic’s library of seizure-related EEG data, it correctly identified nine out of 10 seizures in a pilot study.
Some experts say these models may be useful in specific cases, but doubt that an all-encompassing AI brain model is feasible—particularly one relying on consumer data. It would have to generalize across different kinds of brain states, diseases, and populations—from people with Parkinson’s disease all the way to paralyzed patients wanting to communicate. “It’s still a very, very long way off,” says Adam Caplan, founder and general partner at Jumpspace Ventures, a venture capital firm focusing on neurotechnologies.
He and others say building such a model would require high-quality data that consumer neurotechnology devices often do not collect. Yet nine consumer neurotechnology and AI software companies have publicly acknowledged building a brain foundation model, and others may be doing so quietly.
They face few constraints in exploiting their users’ data. Science reviewed the privacy policies of 57 consumer-oriented companies from around the world that collect EEG data. Four had no privacy policy at all, potentially breaching data privacy laws, whereas 11 had a policy focused just on website visitor data. Among those that did have a broader privacy policy, about two-thirds explicitly mentioned collecting brain data. “That is not just deeply disturbing, it is cause for alarm and corrective action,” Jones says.
Initially, none of the consumer neurotech companies building a foundation model explicitly disclosed this in their privacy policies, although the efforts were disclosed elsewhere, for example in press releases. Emotiv, for example, told Science it trains its foundation model—referred to as its “large brain model”—in part on data from users of the company’s higher end devices. (Emotiv also draws on other sources such as public EEG data sets and partnerships with universities and multinational companies for neuromarketing research.) After Science’s inquiries, the company updated its privacy policy to disclose the use of deidentified data for brain foundation model training subject to opt-in consent from users. “This update represents a forward-looking commitment,” Le says.
Similar concerns arise when companies exploit data from other sources. Piramidal has trained its model on more than 4 million hours of EEG data from open-source data sets and partnerships with hospitals and institutions. “We did these automated searches on the internet in order to find publicly available and curated data that is not proprietary,” says Dimitris Fotis Sakellariou, the company’s founder and CEO.
These include sources such as Temple University’s EEG Corpus, which contains 60,000 recordings, including seizure-related data, and is open to researchers. Participants in such studies consent to providing their data for research, but neuroethicists warn they are unlikely to have envisioned their data being used to train sophisticated AI brain models. “It matters if you agree to something and then there’s a totally different context of use that you never contemplated,” says Nita Farahany, professor of law and philosophy at Duke University.
But Sakalleriou argues people participating in these studies are precisely the kinds of patients who want to see brain foundation models come to life. “I have spent a lot of time in hospitals, and I can tell you for a fact that these people that gave consent for their EEGs are waiting for something to be built like that.”
It’s not just how data are used, but where they might end up that raises concerns. Of those companies with privacy policies, all except one allow data to be shared with third parties in anonymized, aggregated, or deidentified forms, although they often do not specify whether this includes brain data. Only six explicitly stated they would not sell the user’s brain data, although another 17 state in general terms that they don’t sell or trade personal information. All companies reviewed either explicitly permitted data to be shared with authorities—including police, government, or courts—upon request, or did not specify.
Science also found that more than two-thirds of the companies explicitly authorize corporate transfers of data—the rest do not specify. In theory this means if one company were acquired by another, its brain data could be passed on. “We heard from several companies that that is actually what they are aiming for: to build the technology and then finally be acquired,” says Laura Bernáez Timón, neurotechnology fellow with the Centre for Future Generations, a think tank focused on the future of technology. A small company might follow stringent ethical codes and privacy standards as it collects data, only to be acquired by a larger company with fewer scruples and a greater interest in the data themselves, she adds.
The accumulation of our most intimate data by neurotechnology companies is uniquely concerning, says Andrea Lavazza, professor of moral philosophy and director of the Observatory for the Ethics of New Technologies at Pegaso University. “When you start collecting brain data, you’re crossing into the final frontier,” he says. “There’s really nothing beyond that when it comes to our privacy.”
Others are less concerned because of the inherent limitations of consumer-oriented neurotechnologies. They also note that other forms of data collection that we’ve become used to can reveal as much, if not more, about our mental lives. Companies can infer what captures our attention by analyzing our clicks on social media, for example, or even by looking at eye movements picked up through eye-tracking technologies embedded in cameras. By layering multiple kinds of data, they could build up a sophisticated understanding of our thoughts and feelings, without needing to peer into our brain. “We should worry about all of the kinds of data that are being used to make increasingly more precise inferences about mental states,” Farahany says.
Regardless of the risks, experts and many consumers agree that neurotech companies cannot just assume they have users’ consent to use their data for whatever purpose they deem fit. If firms want to train AI models on user data they should get “a specific and separate consent” for that purpose, Farahany says. Science’s privacy policy analysis “reinforces very strongly the need for governments and regulators around the world to understand that this is science and not science fiction,” says Jared Genser, co-founder and general counsel of the Neurorights Foundation.
Some countries have begun to tackle the problem. In 2021, Chile became the first nation to protect brain data within its constitution, requiring special safeguards for brain activity and the information derived from it. The following year, former Chilean senator Guido Girardi sued Emotiv for collecting and using his brain activity without adequate consent. In August 2023, the Chilean Supreme Court ruled in Girardi’s favor, ordering Emotiv to delete all brain data collected from him and directing regulators to review the sale and use of the company’s five-channel Insight device in the country.
More recently, several U.S. states—including Colorado and California—have amended state consumer privacy laws to include neural data, meaning brain data are legally considered sensitive personal information and people can request access to or deletion of this information, among other safeguards. A law introduced by Senator Chuck Schumer (D–NY) would require the Federal Trade Commission to investigate gaps in protections for neural data and propose a regulatory framework for how brain data are collected, shared, used, and sold.
For now, selling brain data appears to be too bold a step for most firms. “Companies recognize that it’s a critical decision point, but I have not heard of a single company who has chosen to monetize or sell the data,” Farahany says. But it’s a regular topic of conversation in company offices, according to several insiders and researchers, including Farahany and Olivier Oullier, neurotechnology entrepreneur and professor of human-computer interaction at the Mohamed bin Zayed University of Artificial Intelligence.
The incentives are substantial. “Neurotech is not cheap,” Oullier says. Bringing neurotechnologies from early-stage research to market requires significant technical expertise and expensive hardware, making a return on investment far from guaranteed.
Some companies are also exploring paying consumers for their brain data, which could then be traded through shared digital databases built on blockchain technology. “Something that we’re going to see more and more of is people in the blockchain and crypto world creating these big data wells,” said Karen Rommelfanger, neuroethicist and director of the Institute of Neuroethics. “They want people to give them their data, want to have access to it, and sell it.”
This is the approach of Chinese company MATRIX, whose setup includes a noninvasive headset equipped with a chip that purportedly samples brain signals 1000 times per second to detect changes in attention, relaxation, and sleep. The company is currently using the headset in pilot studies, says Chief Marketing Technology Officer Eric Choy, but the next milestone is to open a commercial marketplace “where everyday users lease their sleep data and earn income.” The platform—called NeuraMATRIX—will allow users themselves to decide who gets access to their data. This could potentially include medical researchers, pharmaceutical companies, or AI developers.
Such schemes might alleviate concerns about permission and privacy. But they could exacerbate social and economic inequalities, says Philipp Kellmeyer, junior professor for responsible AI and digital health at the University of Mannheim. Vulnerable people might be more inclined to sell their brain data, he says, not unlike individuals who choose to sell their organs. “If people have to monetize their body data, their bio data, in order to keep surviving economically, it would also undermine their human dignity.” MATRIX did not respond to multiple requests for comment on these concerns.
As the neurotechnology industry continues to grow and new ways to trade or sell brain data emerge, experts like Marietjie Botes, a research fellow at the University of Texas at Austin who studies the international governance of brain data, worry regulators and ethicists will struggle to keep pace.
“Technologies move very fast,” she says. In China, for instance, companies are already harvesting hundreds of thousands of hours of EEG data from truck drivers and construction workers—in some cases without anonymization—raising privacy concerns. “Suddenly, from a governance perspective, all the risks we’ve hypothesized—everything is now happening.”
.
Adnan Jinnah started using Interaxon’s Muse headband because he was having trouble falling asleep. MARK PINDER
.
.
Click the link below for the complete article:
.
__________________________________________
Leave a Reply