TL;DR:

1. The IUCN Red List is one of the most significant tools in global conservation, guiding conservation action and policy worldwide.
2. Two key problems faced by the Red List assessments are (1) data gaps and uneven coverage of species, and (2) the amount of time, money and expertise required to produce an assessment and keep them up to date.
3. AI could help solve these problems at scale. By leveraging today’s unprecedented advancements in both NLP (state-of-the-art LLMs that can mine through massive corpuses of relevant literature) and geospatial foundation models (e.g. TESSERA), AI could assist human experts save precious time and resources in managing this crucial conservation tool.

Full Description:

The IUCN Red List of Threatened Species is the world’s most comprehensive inventory of the global conservation status of plant and animal species. It provides a standardised, evidence-based framework for evaluating extinction risk of species worldwide, grouping species into risk categories based on quantitative criteria. It is thus widely relied upon by governments, NGOs, researchers and conservationists to guide and prioritise conservation action and policy.

The IUCN has set an ambitious goal of assessing 260,000 species by 2030, but, as of 28 September 2025, 90,580 further species still require assessment to get there. To reach this goal, and more generally, two major problems are faced in producing and maintaining the Red List:

1. Resource intensiveness
   • Assessments require large amounts of time, financial resources, and expert input.
   • Many assessments become outdated because they aren’t revised frequently enough — some are more than 10 years old.
2. Data gaps and uneven coverage
   • Many species (especially invertebrates, plants, fungi, and microorganisms) are Not Evaluated or Data Deficient.
   • Mammals and birds are much better covered than insects or deep-sea organisms, creating a bias. 84% of the ~75k vertebrate species have been assessed, compared to just 18% of ~425k plant species, 2% of invertebrates, and <1% of fungi and protists.
   • Geographic bias: species in well-studied regions (e.g., Europe, North America) are often better assessed than those in tropical or remote areas where biodiversity is richest.

Today’s ground-breaking advancements in AI could help solve these problems at scale. AI can be used to mine scientific literature, grey literature, and local/indigenous knowledge documents (e.g. local field guides) to find relevant knowledge on data-deficient species. Needle-in-a-haystack tasks like this that could take hundreds of hours of precious time of human experts can be completed by AI in minutes. Moreover, as new literature is released, AI could be used to keep the assessments up to date. AI could also be used agentically, leveraging the latest innovations in geospatial foundation models like TESSERA, to automatically to train and re-train AoH classifiers (by regularly pulling in from observation datasets like iNaturalist) that can inform population estimates and keep estimates and ranges up to date.

This project will thus explore these techniques for leveraging AI to rapidly scale the ability of expert practitioners to assess and maintain knowledge about species conservation statuses, which could be invaluable as we race against time in our fight against the global biodiversity crisis.