A silent sentinel on your wrist, watching for what you cannot yet see
At the University of Campinas, researchers have taught a smartwatch to recognize the quiet language of anxiety before the person wearing it can name what they feel. By weaving together the heart's electrical rhythms and the arm's daily movements into a unique biometric signature, an artificial intelligence achieves over 80% accuracy in detecting stress — not to replace the physician, but to whisper a warning before silence becomes crisis. It is a small device asking a large question: what might we do differently if we knew sooner?
- An AI trained on deliberately induced stress — participants racing to multiply three-digit numbers against a countdown on their own wrist — can now recognize anxiety in the body before the mind catches up.
- The tension lies in the gap between detection and action: the system flags recurring episodes with over 80% accuracy, yet regulatory approval stands between the laboratory and the wrists of real people in the real world.
- Researchers are careful to frame the smartwatch not as a diagnosis but as a sentinel — a proactive layer that recommends a specialist visit while leaving every decision in the user's hands.
- The same Unicamp team is simultaneously fighting a parallel crisis of trust, deploying deepfake detectors and misinformation tools already used by governments and fact-checkers across two continents, arguing that health monitoring and truth verification are ultimately the same problem.
A smartwatch that senses your anxiety before you do is no longer science fiction. At Viva Bem, a research center at the University of Campinas funded by FAPESP and Samsung, a team led by professor Anderson Rocha has developed AI software that detects anxiety with more than 80% accuracy by reading two continuous streams of data from the wrist: an electrocardiogram tracking the heart's electrical activity, and accelerometry mapping the arm's movements throughout the day. Together, these signals form what researchers call a "data signature" — a pattern unique to each person that the algorithm learns to recognize over time.
To teach the AI the difference between calm and anxious states, the team designed clinical protocols that deliberately provoke stress in controlled settings. In one exercise, participants must mentally solve multiplication problems like 309 times 17 in thirty seconds while watching a countdown timer on their own wrist. The physiological response to that pressure becomes the training data, teaching the system to spot the same patterns in ordinary life.
Rocha is deliberate about what the technology is not. It is not a diagnostic tool and is not meant to replace doctors or psychologists. It functions instead as a silent early warning layer — if recurring anxiety episodes are detected, the device recommends consulting a specialist, but the choice of what to do remains entirely with the user. The same philosophy guides the project's expansion into hypertension, diabetes, Parkinson's disease, and fall risk in elderly people. Before any real-world deployment, the team will seek approval from Brazil's health regulatory authority, Anvisa.
At the same London presentation where Rocha unveiled these findings, he also described a parallel project called Horus, which builds tools to detect deepfakes, fraudulent messaging attacks, and fabrications in scientific publications. One solution is already in use by the U.S. Office of Scientific Integrity; others are deployed by Brazilian fact-checking agencies and were used to analyze visual records from recent Middle East conflicts. For Rocha, the two projects share a single foundation: trust. Human-centered AI, he argues, is not about replacing human judgment — it is about giving people better, earlier information so they can make better choices for themselves.
A smartwatch that knows you're anxious before you do. It sounds like science fiction, but researchers at the University of Campinas are making it real. In laboratories at Viva Bem—a research center funded by FAPESP and Samsung—a team has developed artificial intelligence software that detects anxiety with more than 80% accuracy by reading the signals your body sends through a wristwatch.
Anderson Rocha, a Unicamp professor and coordinator of the center, presented these findings in early June at FAPESP Week in London. The technology works by continuously collecting two types of data from the smartwatch: an electrocardiogram that tracks the electrical activity of your heart, and accelerometry that maps the movements of your arm throughout the day. Together, these signals create what researchers call a "data signature"—a unique pattern specific to each person that the AI learns to recognize and monitor over time.
To train the algorithms to tell the difference between calm and anxious states, the team designed clinical protocols that deliberately induce stress in controlled ways. In one test, participants are asked to mentally calculate the answer to multiplication problems like 309 times 17 in just 30 seconds while watching a countdown timer on their own wrist. The anxiety that inevitably follows becomes the teaching moment. Researchers measure how the body responds to this exercise and use that data to train the algorithms to spot the same patterns in everyday life.
Rocha is careful to emphasize what this technology is not. It is not meant to replace doctors or psychologists. The smartwatch is not a diagnostic tool. Instead, it functions as a proactive monitoring layer—a silent sentinel on your wrist. If the device detects recurring episodes of anxiety, it sends an alert recommending that you consult a specialist. The decision about what to do with that information remains entirely with the user. The same principle applies to other conditions the project is exploring: hypertension, diabetes, Parkinson's disease, and the risk of falls in elderly people. In each case, the AI serves as an early warning system, not a verdict.
"The goal is to identify the first symptoms of different health conditions using the signals captured by smartwatches," Rocha explained, "so we can help people have a better quality of life." The results are still being evaluated and continuously refined. Once the team considers the technology mature enough, they will seek approval from regulatory authorities like Brazil's National Health Surveillance Agency (Anvisa) to begin testing with real users in the real world.
At the same London presentation, Rocha also discussed a parallel project called Horus, which tackles what his team calls "synthetic realities"—the growing universe of images, videos, and text generated by artificial intelligence. The lab has already built tools to detect deepfakes, to identify fraudulent SMS and WhatsApp attacks, and to spot fabrications in biomedical scientific publications. One of these solutions is already in use by the U.S. government's Office of Scientific Integrity and is available as open-source software. Another tool for verifying images is being used by fact-checking agencies in Brazil, including Lupa, Aos Fatos, and G1, and was deployed to analyze visual records from recent conflicts in the Middle East—cases that made their way into reports by Reuters and Agence France-Presse.
For Rocha, health monitoring and fighting misinformation are not separate problems. They converge on a single value: trust. "Human-centered AI is fundamental to strengthening resilience and well-being," he said. The smartwatch project represents one piece of a larger vision—technology that serves people by giving them better information about themselves, earlier, so they can make better choices about their own care.
Citações Notáveis
The goal is to identify the first symptoms of different health conditions using smartwatch signals so we can help people have a better quality of life— Anderson Rocha, Unicamp professor and coordinator of Viva Bem research center
Human-centered AI is fundamental to strengthening resilience and well-being— Anderson Rocha
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that the watch detects anxiety before the person feels it? Isn't that a bit intrusive?
It's not about reading minds—it's about pattern recognition. Your body shows signs of stress before your conscious mind catches up. The watch is just making visible what's already happening. If you're having recurring anxiety episodes, knowing about them earlier means you can seek help sooner, before they compound.
But how do you teach a machine to recognize something as subjective as anxiety? Anxiety feels different for different people.
That's exactly why they developed those clinical stress tests. By putting people through the same controlled challenge—the mental math problem—they create a moment where anxiety is real and measurable. The heart rate spikes, the movement patterns change. The AI learns those physical signatures, then watches for them in daily life. It's not judging whether you're anxious; it's recognizing your body's particular way of showing stress.
What happens when the alert comes? Does the person have to do anything?
Nothing mandatory. The watch sends a recommendation to see a specialist. But the person decides. That's the whole point—it's a tool, not a diagnosis. Rocha was very clear about that. The technology doesn't replace doctors; it just gives people information they might not have had otherwise.
And this is just the beginning. What else could it detect?
They're already looking at hypertension, diabetes, Parkinson's, fall risks in elderly people. Anything where the body gives off signals before symptoms become obvious. The real question is whether people will trust it, and whether regulators will approve it for actual use. That's still ahead.
So we're not there yet.
Not quite. The lab results are strong—over 80% accuracy. But real-world testing with actual users hasn't happened. They need regulatory approval first. That's the next hurdle.