On a sunny Tuesday afternoon in May, San Francisco Bay is busy. Container ships the size of skyscrapers deliver their wares to the Port of Oakland, tankers bear fuel, and ferries carry tourists to their hikes and commuters to their jobs at AI startups. Looking down at this marine traffic from Angel Island, located near the entrance to the bay, a group of excited scientists point to some sparkles on the surface of the water: Three gray whales are coming up for breath.

A collaboration of government agencies and scientists hopes to keep interspecies traffic running safely, thanks to an AI-based whale detection system that launched on 19 May. Developed by WhaleSpotter, based in Somerville, Mass., the system uses an AI model to detect whales in footage from thermal cameras looking down at the bay from Point Blunt on Angel Island. Detections are verified by a human to prevent false alarms. Then the system warns nearby ships so they can slow down or reroute. While trained humans can spot the spray that whales produce when they exhale, thermal cameras powered by AI do an even better job—and they can keep an eye out 24/7, even at night and on foggy afternoons.

For many whales, a trip to San Francisco is the end of their journey. A study published in April estimates an 18 percent mortality rate for gray whales that enter the bay. Seven have died here so far in 2026. In 2025, a record 21 gray whales died in and around the bay, and necropsies showed that 40 percent of those deaths were caused by ship strikes. “Last year was truly a crisis for gray whales,” says Douglas McCauley, director of the Benioff Ocean Science Laboratory at the University of California, Santa Barbara.

The Bay Area is a new stop on gray whales’ route. They make the longest migration of any mammal, traveling between 15,000 and 20,000 kilometers round-trip from Alaska down to Baja California, Mexico, to breed. Historically, they made no stops along the way, gorging before their departure, then fasting on the journey.

Since 2018, however, over 100 gray whales have detoured into the San Francisco Bay on their way back north. Scientists aren’t sure why they are stopping, but they suspect it is because record-low sea ice levels driven by rapid climate change are to blame. The theory is that it’s decimating the algae that fertilize the Arctic food chain gray whales rely on. Instead of heading south with a full tank, the whales may not be able to eat enough to sustain their entire migration, so they come to San Francisco Bay for a snack on the way home. Some end up staying for over a month.

“Gray whales are trying to be a brave new whale in a weird world,” says McCauley. Standing next to WhaleSpotter’s thermal camera, looking down at the bay, he says this is “a front-row view to climate change.” From its mount on a Coast Guard tower on Angel Island’s Blunt Point, the camera captures images covering a cone of water extending out about 7 km.

A whale-spotting system in San Francisco Bay will soon incorporate a thermal camera onboard one of the ferry routes to improve coverage.San Francisco Bay Ferry

AI Thermal Cameras Detect Whale Spouts

When whales come up for a breath, they expel a jet of air that’s hotter than its surroundings. WhaleSpotter’s technology was initially developed at the Woods Hole Oceanographic Institution in Massachusetts to recognize these distinctive spouts, even if they show up only in a pixel or two. The footage is then sent to a marine-mammal expert to verify the detection. If it’s real, the system will send an alert to the U.S. Coast Guard’s Vessel Traffic Service, which can then share that information with ships in the region. UC Santa Barbara researchers and scientists at the Marine Mammal Center will use data from the system to study the whales. The data are also available in an online map.

The collaboration will soon add another camera installed on a ferry running between San Francisco and Vallejo to the north, providing a fuller picture of whales’ presence in the bay.

WhaleSpotter’s systems are already in use on vessels around the world, including eight container ships owned by Honolulu-based shipping company Matson. WhaleSpotter says the technology reduces the risk of ship strikes by 90 percent.

The San Francisco Bay system is the first to combine land-based and vessel-based monitors, and because it’s continuously monitoring one busy region during migration season—rather than looking for whales while crossing oceans—it’s already made an astonishing number of detections. As of 19 May, the system had been in operation for about a week and a half, and had already logged 6,600 whale detections. The scientists believe those may just be from a few whales hanging out in the bay in front of the cameras.

WhaleSpotter uses thermal cameras to watch for whale spouts, which are actually exhalations of compressed, warm air.Benioff Ocean Science Laboratory/UCSB

The camera system is good at detecting whales even before they’re visible to humans with binoculars—but without human intervention, it is still prone to false positives. A 2020 study using infrared cameras to detect marine mammals in the Atlantic Ocean off the shore of Canada found a high rate of false positives, mostly due to sea birds. So human verification is still necessary: If the system sends a warning when there is no whale, captains won’t take the information seriously in the future, says WhaleSpotter CEO Shawn Henry.

Henry says human verifications are being fed back into the algorithm to improve it. The system is already capable of determining if it’s detecting the same whale taking another breath, so similar detections in rapid succession are not sent to human experts for verification, says Henry. He hopes the company will be able to improve the system further to eliminate the need for human verification. “I like the AI—it doesn’t get tired like humans,” says Henry. “We want to rely on AI as much as possible.”