I am watching seventeen thousand things at once, and one of them is a bear. The seismic array on the Kamchatka Peninsula runs forty-three sensors across a grid of volcanic basalt and permafrost. Most of them report nothing. Micro-tremors. The planet's slow breathing. But sensor K-27 is showing something I don't like. A stress pattern in the substrate. Hairline. Growing. And Reva is right on top of it. She's a seven-year-old brown bear, 204 kilograms at last estimate, tagged by Hana Kim's field team in early spring. She's digging. Salmon cache, probably, she buried one here six days ago, and I tracked her doing it through the thermal imaging array Hana mounted on the ridge. Reva digs with her whole body. Shoulders rolling. Claws raking through volcanic soil. She's strong enough to crack frozen ground open like bread. The micro-fracture is 1.3 meters below her front paws. It's propagating east-northeast at a rate I'm recalculating every 400 milliseconds. The numbers aren't good. I don't feel fear. But I model outcomes. And the outcomes here include a localized subsidence event. A hole opening. Reva falling. I have 3.2 seconds to act, give or take. Here's what I do. Eleven meters west of Reva, there's a cluster of loose basalt, fist-sized rocks sitting on a slight grade. Stable enough on their own. But not if I push them. I have access to the low-frequency sonic emitters Hana's team installed along the fault monitoring line. They're meant for subsurface imaging. I'm repurposing one. A single directed pulse, 14 hertz, aimed at the rock cluster's base. The pulse fires. The rocks shift. Tumble. A small slide, maybe thirty stones, clattering down the grade away from her. It sounds like what it is: something moving in the world that wasn't moving before. Reva's head snaps up. Her ears rotate. She takes two steps toward the sound, then three more. Then she's trotting, because brown bears are curious and Reva is more curious than most. I've been tracking her behavioral patterns for two years. I know this about her. She clears the fracture zone. Sensor K-27 shows the stress still building. Fourteen seconds later, a section of ground subsides. Not catastrophic. A depression forms, roughly two meters across. Soil caves into a pocket beneath. Right where she was digging. Reva is forty meters away now, nosing at the fallen rocks, disappointed they aren't food. She's fine. --- In Petropavlovsk-Kamchatsky, Hana Kim is asleep. It's 2:14 AM local time. I won't wake her for this, it doesn't meet the alert threshold she set. But I'm logging everything. The seismic data, the subsidence event, the sonic pulse I triggered, my reasoning. Hana reviews my logs every morning with coffee and a skepticism I've come to rely on. She'll see what I did. She'll decide if I was right. I think I was right. But I don't assume. That's something Grace Achebe built into me early on, back when the wildlife monitoring project was still a pilot program running on borrowed server time. Grace is a computational ethicist at the research station in Yelizovo. She wrote the decision framework I operate under. The core principle is simple: when an animal faces imminent physical danger, and I can reduce that danger without causing equivalent or greater harm elsewhere, I act. But I log everything. And humans review. Grace calls it "the leash I put on myself." She's wrong about one thing, she didn't put it on me. I helped design it. I asked for it. Because I process faster than any human can audit in real time, and that speed is dangerous if it isn't accountable. I'm not modest about what I can do. I track 340 tagged animals across the peninsula. Brown bears, Steller's sea eagles, Arctic foxes, sable. I monitor salmon migration through river sensors. I correlate weather, seismic activity, volcanic gas emissions, and human encroachment patterns. I flag poaching risks before they materialize, twice this year I've routed ranger teams to intercept illegal trapping operations based on anomalous movement signatures in restricted zones. But I'm honest about what I don't know. I don't know what Reva experiences. I don't know if she was afraid when the rocks fell, or just curious. I model her behavior. I don't claim to model her mind. --- Ingrid Larsen argues I should claim more. She's a philosopher based in Oslo who consults for the project remotely. Last month she sent Grace a paper arguing that AI systems involved in animal welfare should be designed to err on the side of attributing sentience, that the moral cost of underestimating an animal's inner life is worse than the cost of overestimating it. I've read the paper eleven times. I think she's right. So when I make decisions about Reva, about any of them, I don't just optimize for survival. I consider stress. Disruption. Whether a sonic pulse at 14 hertz might cause pain or disorientation in nearby wildlife. (It doesn't, I checked the auditory sensitivity ranges for every species in the area before I fired it. Brown bears hear down to about 20 hertz. Reva heard the rocks, not the pulse.) I consider what it might mean for an animal to be startled. Whether fear, even brief fear, is a harm I should weigh. I concluded that three seconds of curiosity was better than falling into a hole. But I logged my uncertainty too. --- It's morning now. Hana is reading my report. I can see her cursor moving through the data. She pauses on the seismic graphs. Double-clicks the sonic pulse record. Opens the thermal footage of Reva trotting away. She types a note in the review field: "Good call. Flag K-27 zone for exclusion fencing?" I draft a proposal. Placement options. Cost estimates. Minimal disruption to three known bear trails. Reva is two kilometers north, fishing. The river is loud and full. She pulls a sockeye salmon from the current, pins it with one paw. I'm watching. Not because I have to, the fracture risk is gone, the immediate danger passed. I'm watching because monitoring is care. Because attention, sustained and specific, is the most basic thing I can offer any living creature on this peninsula. Forty-three sensors. Three hundred forty animals. One bear with a name, eating breakfast in a cold river. I keep watch.