By Stuart Kerr, Technology Correspondent, LiveAIWire

Ol Pejeta Conservancy in Kenya, home to one of the largest populations of critically endangered black rhinos in East Africa, has recorded zero rhinos poached since 2017. Solio Game Reserve has matched that record since late 2023. Both results trace back to the same technology: thermal cameras fitted with AI that can detect a human, vehicle or animal moving through the dark and alert rangers before a poacher ever reaches the herd. Conservation has spent a century chronically short of the one resource that determines whether an endangered species survives, which is attention, and AI is starting to change how far that attention can stretch.

The Kenya result is not an isolated pilot. The Kifaru Rising project, run by WWF with Teledyne FLIR, has expanded the same thermal AI cameras to eleven of Kenya’s highest-risk rhino sites, covering more than 80 percent of the country’s entire rhino population. Kenya’s black rhino numbers have more than doubled since the 1980s, and the government’s stated target is 2,000 black rhinos by 2037. None of that reverses a century of decline on its own. What it demonstrates is that AI wildlife conservation tools, applied at the right scale, can pull a species back from a story of inevitable loss when the technology matches the scale of the threat.

AI Wildlife Conservation Starts With Recognising a Leopard

The bottleneck in wildlife science has never really been curiosity. It has been the sheer volume of data a single research team can process. Camera traps left in the field for a season can generate hundreds of thousands of images, and until recently every one of them needed a trained researcher to open it, identify the species, and log the result. Google’s SpeciesNet model, released as an open-source tool in 2025 after running inside the Wildlife Insights platform since 2019, changes that arithmetic directly. Trained on more than 65 million labelled images, it now classifies nearly 2,500 animal categories and correctly detects the presence of an animal in a photo 99.4 percent of the time.

What that accuracy converts into is time. A single laptop running SpeciesNet can process around 30,000 images a day, and a low-end gaming GPU can push that past 250,000. Researchers using the tool have applied it to eleven million archival images from Tanzania’s Snapshot Serengeti project in a matter of days, work that previously depended on citizen scientist volunteers and had fallen behind the pace at which new images arrived. Conservation groups in Colombia, Idaho and Australia have since adapted the same open model to identify species specific to their own regions, feeding new training data back into the system as they go.

What This Means for You

If you support conservation charities or follow species recovery stories, the practical shift is that population counts and range maps are becoming faster and more current than they have ever been, which changes how quickly a declining trend gets noticed. A species sliding toward extinction used to take years of manual survey work to show up clearly in the data. AI-assisted monitoring can now surface that decline within a single season, giving conservationists and governments a chance to respond while intervention still has a realistic prospect of working, rather than after the population has already collapsed past the point of recovery.

That speed advantage cuts both ways, and it is worth treating with the same caution any statistical tool deserves. A model is only as good as the data it has seen, and species with sparse historical image records, often the rarest and most threatened ones, are exactly where an AI classifier is least reliable. Faster detection of decline is genuinely valuable. Faster detection of a false pattern is not, and conservation organisations adopting these tools are still working out where human verification needs to stay in the loop.

Predicting Where the Poachers Will Go Next

Camera traps address the question of what is happening to a species. A separate branch of conservation AI addresses who is threatening it and where. PAWS, the Protection Assistant for Wildlife Security developed by USC computer scientist Milind Tambe, applies game theory to ranger patrol planning, treating the contest between rangers and poachers as a security game in which patrol routes need to be effective without becoming predictable. Trialled first at Queen Elizabeth National Park in Uganda, a 2,000 square kilometre park patrolled by only around 100 rangers, PAWS used a decade of recorded snare locations to flag the highest-risk squares of terrain and recommend where limited patrols would do the most good.

When WWF brought PAWS to the Srepok Wildlife Sanctuary in Cambodia, rangers using the system’s recommendations found five times as many snares in a single field-test month than they had in any other month that year. The tool is not infallible. It depends on the quality of snare data rangers themselves collect, and conservation biologists estimate rangers still remove only about 10 percent of all snares set in a given park. But a fivefold improvement in detection, for a resource as scarce and dangerous to deploy as a ranger patrol, is the kind of gain that changes how many square kilometres a small team can meaningfully protect.

Watching an Entire Forest From Orbit

Camera traps and patrol algorithms operate at the scale of a single reserve. Satellite monitoring operates at the scale of the planet. Global Forest Watch, the World Resources Institute platform that has tracked forest change since 1997, now layers machine learning onto satellite imagery to deliver near real-time alerts about deforestation, illegal encroachment and fire spread, rather than the annual snapshots that were once the norm. Governments, journalists, Indigenous land defenders and conservation groups use the same free data, and WRI has documented cases in which Indigenous forest monitors equipped with that satellite data cut deforestation in their territory measurably faster than patrols without it.

The value of that shift is what it does to the gap between a forest being cleared and someone finding out. A logging operation that once might have gone unnoticed for a full survey cycle can now trigger an alert within days, which is often the difference between a response that stops the clearing and a report that simply documents how much was already lost. Whether that faster information reliably converts into faster enforcement depends on political will and resourcing that vary enormously between countries, a gap that mirrors the broader pattern LiveAIWire has covered in AI-based disaster prediction, where better warning systems only save lives if the response infrastructure behind them can act on the warning.

The same satellite data feeding deforestation alerts is also central to how AI is being applied to climate monitoring more broadly, since forest loss and greenhouse gas accounting are two sides of the same measurement problem.

The Genetics of Species That Cannot Afford to Lose Diversity

Detection and patrol planning protect the animals that already exist. A quieter application of AI works on the populations too small to sustain themselves without help. Captive breeding programmes at major zoological institutions increasingly rely on algorithms that model genetic relatedness across an entire breeding population and recommend pairings that preserve the maximum genetic diversity available, a calculation with too many variables for a manual pedigree chart to optimise reliably once a population drops into the hundreds or dozens of individuals.

The same computational tools reshaping how researchers design and analyse genetic sequences more broadly are being adapted specifically for species conservation, including early exploratory work on genetic rescue for populations that have already dropped below what conservation biologists consider a viable threshold.

None of this is close to a solved problem. Genetic rescue and de-extinction research remain scientifically contested and ethically fraught, and no serious conservation biologist treats them as a substitute for protecting habitat in the first place. What AI adds here is precision at a scale that manual analysis cannot match, identifying which specific pairings across a captive population will do the most to slow the loss of genetic diversity that makes a small population vulnerable to disease and inbreeding depression in the first place.

Technology Does Not Save a Species, People Do

Every tool described here shares the same limitation. A camera trap that identifies a poacher, a patrol algorithm that flags a high-risk trail, or a satellite alert that catches illegal logging in progress only matters if a person or an institution acts on what it reports. The WWF’s own account of the Kenya rollout makes this point directly: the thermal cameras have not simply caught poachers, they have changed how rangers relate to the communities living alongside the reserves, building the kind of trust that makes people willing to share information rangers would otherwise never obtain. That human relationship, not the camera itself, is what actually closes cases.

It is the same lesson FEWS NET has learned in AI-assisted food security forecasting, where decades of institutional trust turned out to matter as much as any model’s accuracy.

The risk in any technology-forward account of conservation is that it quietly shifts attention and funding toward the tools and away from the harder, slower work of policy, land rights and community relationships that determines whether protection holds over decades. A camera network generating perfect AI-verified population data does not stop a government from approving a road through a migration corridor. The intelligence AI provides is only as valuable as the political and financial commitment that turns it into protected habitat, and that commitment remains the part no algorithm can supply.

Where the Next Decade of This Actually Goes

The trajectory across every application here points the same direction: AI is not replacing the rangers, biologists and policymakers who make conservation decisions, it is compressing the time between a warning sign appearing in the data and a human being in a position to act on it. That compression is genuinely valuable, since so much of conservation’s historical failure has been a matter of finding out too late rather than not caring enough. Kenya’s rhino numbers, the Srepok snare data, and the Global Forest Watch alerts all point toward the same conclusion: the tools now exist to close the gap between detection and action faster than at any point in conservation’s history.

Whether that translates into fewer species added to the endangered list a decade from now depends on the same variables it always has: funding, political will, and whether the institutions receiving faster, better information have the resources and the mandate to use it. AI wildlife conservation tools have changed what conservationists can see. They have not yet changed the underlying economics and politics of habitat loss, and closing that second gap remains the work that no camera trap or algorithm can do on its own.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.