Kew Gardens scientists have used artificial intelligence to assess extinction risk across every known flowering plant species for the first time. The results: Kew's State of the World's Plants and Fungi Report estimates 45% of flowering plant species are at risk.

That number matters because flowering plants (botanically called angiosperms) represent roughly 90% of all plant life on Earth. They form the foundation of nearly every terrestrial food chain, support most agricultural systems, and provide the majority of medical compounds now in clinical use. When nearly half are under threat, the problem goes beyond biodiversity loss—it raises questions about the stability of the systems that feed and heal us.

Conventionally, the International Union for Conservation of Nature (IUCN) assesses species one by one, a process that demands taxonomic expertise, field surveys, and years of review per species. That bottleneck has left most plant species never formally evaluated. Kew scientists trained their AI models on the fraction of species already assessed, then projected those patterns across all flowering plants, generating the first comprehensive risk profile published in March 2024. The output is probabilistic rather than a definitive judgment on each species, but at this scale, probabilistic coverage is exactly what conservation triage requires—directing effort where it is needed most.

The problem is compounded by how much we do not know. In 2025 alone, Kew recorded 125 new plant species and 65 new fungi named to science. That incomplete inventory matters more than it sounds. A 2023 Kew analysis found that three out of four plant species still undescribed are already likely at risk even before they are formally named. The traditional narrative—discover a new species, confirm its safety—has inverted. Finding a species no longer means it is safe, or even abundant.

Geography shapes the problem unevenly. Kew has identified 32 global plant diversity 'darkspots'—regions where many plant species exist but scientific data are scarce. These zones carry the widest uncertainty for the AI model and offer the greatest payoff for targeted field surveys. Many overlap with tropical and subtropical areas experiencing rapid land conversion, meaning the knowledge gap and the actual threat reinforce each other geographically.

What this AI assessment does and does not do matters for how conservation planners use it. It is not a replacement for ground-level threat mapping or the legal frameworks tied to formal IUCN listings. But as a prioritization tool—directing limited resources toward plant groups and regions where risk appears elevated—it fills a gap that has existed since modern conservation science began. The 45% figure should be read as an upper-bound estimate under current conditions, not a final tally; the underlying confidence intervals deserve scrutiny.

Conservation practitioners targeting the Convention on Biological Diversity's Kunming-Montreal targets—particularly Target 4, which aims to stop human-caused extinctions of known threatened species by 2030—now have a new data layer for their national biodiversity plans. Yet Target 4 has an inherent flaw: it focuses on "known threatened species," which leaves undescribed and unevaluated plants outside the frame. If three-quarters of unnamed species are already at risk, and we are adding only a few hundred species to the formal record each year, the policy targets chase a baseline that keeps shifting.

Here is the central tension this research exposes: conservation depends on complete inventories, yet the inventory is proven incomplete just as the species we have identified are deteriorating. The AI methodology does not solve that tension. What it does is make the shape of the problem visible in a way it was not before.