The water is statistically more dangerous that day. Take precautions.
Where the ocean meets human ambition, a former crime-and-cancer data modeler has turned his curiosity toward the sea, asking whether the same pattern-recognition tools that map human risk might also map the ancient rhythms of sharks. Safewaters.AI, born from a chance news alert and five years of quiet development, now claims to identify dangerous conditions with 89% accuracy — arriving at a moment when Australia's waters have already claimed six lives in 2026. It does not promise certainty, only clarity: a map of conditions, not a prophecy of outcomes. The deeper question it raises is not whether we can predict nature, but whether we are wise enough to listen when we can.
- Six people have died in Australian shark attacks by late May 2026, including a 12-year-old boy in Sydney Harbor, making the human cost of inaction impossible to ignore.
- The app correctly flagged Manly Beach and surrounding Sydney waters as high-risk on the exact days four bull sharks struck within 48 hours — a validation that rattled even its own creator.
- A risk score of 70 doesn't mean a 70% chance of attack — it means 70% of known danger patterns are present, a crucial distinction that shapes how the tool must be communicated to the public.
- Valenti is threading a careful needle: building trust in AI-driven ocean safety while resisting the temptation to overstate what pattern recognition can and cannot promise.
- The long-term vision is integration into mainstream weather and surf apps globally — making shark risk as routine to check as wave height or UV index before entering the water.
Evan Valenti was learning to surf in Narragansett when a news alert told him eight great whites had just been tagged in the water he'd been swimming in. At the time, he was building AI models in New York that predicted crime and cancer trends. The question arrived almost on its own: could the same statistical machinery find patterns in shark behavior?
Five years later, that question became Safewaters.AI — an app claiming 89% accuracy in forecasting shark attack risk. Its arrival feels less like coincidence than necessity. Australia's 2026 season has brought nineteen attacks by late May, six fatal. In January, four bull sharks struck Sydney waters within 48 hours. A twelve-year-old died in the harbor. A surfer at Long Reef was hit. Hours later, a twenty-seven-year-old at Manly suffered life-changing injuries. Valenti checked his database: those beaches had been flagged high-risk on those exact days.
The engine behind it is a neural network — not a language model, but one designed to hunt for relationships. It ingests over thirty marine variables: water temperature, visibility, chlorophyll levels, upwelling events. It cross-references these against more than sixteen hundred shark attack records spanning two centuries, learning which combinations of conditions preceded attacks. When those patterns realign, it assigns a risk score from zero to one hundred.
Valenti is precise about what that score means. A seventy doesn't signal a seventy percent chance of attack — it means seventy percent of the recognized danger signatures are present. The water is statistically more dangerous. The app translates this into low, moderate, and high categories, giving surfers and divers something actionable rather than paralyzing.
Beyond shark forecasting, Valenti has layered in surf reports, NASA satellite chlorophyll data, and tide charts — building toward something like Surfline, but with seven-day shark activity forecasts embedded. At ten dollars a year for advanced features, he's still refining the model. His larger ambition is for shark risk to become a standard layer in every weather and ocean app worldwide. He frames his purpose simply: if it saves one life in a year, it was worth building. The accuracy surprised him. Whether the world will trust it enough to use it remains the open question.
Evan Valenti was learning to surf in Narragansett when he saw a news alert that stopped him cold: eight great white sharks had been tagged in the exact waters where he'd just been in the ocean. He was working in New York at the time, building AI models that predicted crime and cancer trends. The thought struck him then—what if you could forecast shark attacks the way you forecast crime? What if the same statistical machinery that found patterns in human behavior could find patterns in shark behavior?
That curiosity, five years ago, became Safewaters.AI, an app that now claims to predict shark attack risk with 89 percent accuracy. The timing of its arrival feels less like coincidence and more like necessity. In Australia alone, 2026 has brought nineteen shark attacks by late May, six of them fatal. In January, four bull sharks struck within forty-eight hours in Sydney. A twelve-year-old boy died in the harbor. A surfer named Mercury Psillakis was hit at Long Reef. Hours later, another attack at nearby Manly left a twenty-seven-year-old with injuries described as life-changing. When Valenti checked his database, Manly and the surrounding beaches had been flagged as high-risk on those exact days. The app had seen it coming.
The mechanism is not magic, though it requires some explanation. Valenti built a neural network—not a language model like ChatGPT, but a different kind of AI altogether—that hunts for relationships between things. It ingests more than thirty marine weather variables: water temperature, visibility, chlorophyll levels, upwelling events. It cross-references these against two centuries of shark attack records, more than sixteen hundred incidents in total. The network learns which combinations of conditions preceded attacks. When those same conditions align again, it assigns a risk score from zero to one hundred.
But here's what Valenti is careful to clarify: a score of seventy does not mean there's a seventy percent chance of an attack. Think of it differently. Seventy means seventy percent of the recognized danger patterns are present. The water is statistically more dangerous that day. Take precautions. The app translates this into three categories—low, moderate, high—so a surfer or diver can make an informed choice about whether to enter the water.
Valenti tested his model against historical data and found it correctly identified high-risk days in eighty-nine percent of cases. He's also had real-world validation. The app flagged Boca Grande as dangerous on the day a shark attack occurred there last June. It caught the Sydney attacks in January. These aren't predictions in the fortune-teller sense. They're pattern recognition at scale, the kind of thing neural networks do well when fed enough data and the right variables.
To make the app useful beyond just shark forecasting, Valenti added surf forecasts and visibility metrics derived from NASA satellite chlorophyll data, plus tide charts. His vision is to build something like Surfline—the dominant app for checking wave conditions—but with seven-day shark activity forecasts built in. For now, he's charging ten dollars a year for advanced forecasts. He's still refining the model, adding more inputs, testing new variables.
His larger ambition is to see shark forecasting become a standard feature across all weather and ocean apps, accessible and trusted by beachgoers worldwide. The odds of a shark attack remain statistically low. But Valenti frames his goal simply: if the technology saves one life or prevents one serious injury in a year, it's worth doing. He built it out of curiosity about what AI could do. The accuracy surprised him. The question now is whether the rest of the world will trust it enough to use it.
Notable Quotes
We aren't wizards. It's really just telling you the variables line up today, and the waters, statistically, will be more dangerous, so just take proper precaution.— Evan Valenti, developer of Safewaters.AI
Even if we just save one life or limb a year, that's a win in my books.— Evan Valenti
The Hearth Conversation Another angle on the story
When you say the app predicted those Sydney attacks, what exactly did it see that day that made it flag Manly as dangerous?
It saw a constellation of conditions. The water temperature, the visibility, the chlorophyll levels, the upwelling patterns—all of them aligned in a way that matched historical patterns from days when attacks had happened before. The neural network doesn't know why sharks behave differently on those days. It just knows that when these thirty variables line up this way, attacks have been more common.
So it's not predicting shark behavior at all. It's predicting human vulnerability.
That's a sharp way to put it. Yes. It's saying the conditions are right for an encounter to be more likely. Whether that's because sharks are more active, or more hungry, or the water clarity brings them closer to swimmers—the model doesn't care. It just knows the risk profile.
And the eighty-nine percent accuracy—that's tested against two hundred years of data?
Sixteen hundred individual attacks across two centuries. The model correctly identified which days were high-risk in eighty-nine percent of cases. That's a strong signal, but it's not perfect. There will be false alarms. There will be days it misses.
What happens when someone ignores a high-risk warning and nothing happens?
They might stop trusting the app. That's the real challenge. People need to understand it's not a guarantee. It's a statistical statement about conditions. The ocean is always dangerous. This just tells you when it's more dangerous than usual.
Why do you think Valenti is giving this away so cheaply—ten dollars a year?
Because he's not trying to get rich. He's trying to save lives. And he knows that if it's too expensive, people won't use it, and then it can't save anyone.