A two-week research cruise off the coast of Brazil has returned with 31 animals that appear to be species never formally described by science, most of them transparent or gelatinous creatures pulled from the ocean’s midnight zone. The expedition, known as Designing the Future 3, ran from April 15 through April 30, 2026, in the Southwestern Atlantic under the direction of chief scientist Karen Osborn. The finds highlight how little is known about deep pelagic life in the South Atlantic and raise questions about whether new imaging and AI tools can speed up the slow process of turning a first sighting into an official species description.
Why these ghostly deep-sea finds demand attention now
Traditional midwater surveys rely on trawl nets that shred soft-bodied organisms on contact. Gelatinous animals, mucus-feeding larvaceans, and other fragile forms often arrive on deck as unidentifiable fragments, which means entire communities of life have gone uncounted for decades. The 31 new animals documented during this cruise were recorded largely through non-invasive optical systems rather than physical capture, a shift that changes what researchers can actually see and preserve from the deep ocean.
The expedition deployed DeepPIV, a laser-sheet and optics system that creates three-dimensional reconstructions of fragile gelatinous and mucus structures in midwater, including the delicate feeding houses built by larvaceans. That capability, validated in a peer-reviewed Nature study, means structures that dissolve on contact with a net can now be mapped in place. Alongside DeepPIV, the team used EyeRIS cameras and the URI RAD2 origami robot, a soft-bodied sampling device designed to enclose animals without crushing them.
One hypothesis circulating among ocean technologists is that pairing shipboard open-source microscopes with real-time AI annotation could cut the median time from first observation to formal species description by 40 percent or more compared with traditional net-based surveys. No published dataset from this specific expedition yet confirms that figure, and the claim remains untested in peer-reviewed literature. But the toolkit assembled for the cruise, which included AI pipelines built on the FathomNet database and the FathomVerse annotation platform, was explicitly designed to compress that timeline by processing thousands of video frames while the ship was still at sea.
Imaging and AI tools behind the 31 new species candidates
Karen Osborn, who holds an affiliation with the Smithsonian National Museum of Natural History, led a team that combined instruments from at least three institutions. MBARI supplied the core underwater imaging suite. Stanford University’s Prakash Lab contributed two open-source instruments: Squid, a modular confocal microscope capable of resolving cellular-level detail aboard a moving vessel, and the Gravity Machine, a hydrodynamic treadmill microscope that lets researchers observe live swimming behavior under simulated natural flow conditions. The University of Rhode Island provided the RAD2 origami robot for gentle specimen collection at depth.
The integration of these tools matters because each one addresses a different bottleneck in deep-sea taxonomy. DeepPIV captures external and internal three-dimensional structure without physical contact. Squid allows onboard histology-grade imaging that would normally require a shore-based lab weeks later. The Gravity Machine keeps living specimens in motion so scientists can study locomotion and feeding, behaviors that vanish the moment an animal is fixed in preservative. And the FathomNet and FathomVerse AI systems help annotate and cross-reference imagery against a growing library of known deep-sea organisms, flagging potential unknowns for closer inspection.
The result, according to MBARI’s reporting, was a workflow that moved from live encounter to preliminary identification far faster than a conventional cruise could manage. Researchers could tag candidate novelties within hours, rather than waiting for weeks of post-cruise video review. Whether that speed translates into faster formal descriptions, which require peer review, type specimen designation, and journal publication, is the central open question.
What the expedition has not yet proven about deep-sea discovery
Several gaps separate the headline count from confirmed science. No primary species list, voucher images, or taxonomic descriptions from the cruise have been released publicly. The exact identities of the 31 animals, their phyla, genera, and the criteria used to distinguish them from known species, have not been detailed in any accessible document. Without that information, the count remains a preliminary field estimate rather than a verified tally.
Direct statements or logs showing how many specimens were physically collected versus observed only through imaging are also unavailable. That distinction matters because formal species descriptions in zoology typically require a deposited physical type specimen, not just video or photographs. If most of the 31 candidates were recorded only optically, the path to formal naming could be longer and more contentious than the expedition’s technology-forward approach might suggest.
Quantitative performance data from the FathomNet AI pipeline during this specific cruise have likewise not been released. Key metrics such as precision, recall, and the rate of false positives when flagging “unknown” morphologies remain undisclosed. Without those numbers, it is difficult to assess how many of the 31 candidates emerged from robust AI-assisted triage versus more subjective human impressions formed while scanning live video feeds.
There is also the unresolved issue of how to treat structures, rather than whole organisms, in counts of novelty. DeepPIV can reveal intricate mucus houses built by larvaceans and other gelatinous scaffolds that have never been documented in three dimensions before. Some technologists argue that these engineered habitats, which shape nutrient flows and particle capture in the water column, should be cataloged and potentially named in their own right. Others maintain that taxonomic priority must remain on the animals themselves, with houses and webs treated as extensions of known species until strong evidence suggests otherwise. The current expedition’s tally does not clearly state how such structures were handled.
How new tools could change the pace of naming life
Despite those caveats, the cruise offers a glimpse of how deep-sea discovery might work in the near future. Instead of shipping jars of damaged specimens back to shore and waiting months for specialists to examine them, teams can now combine in situ imaging, gentle capture, and onboard microscopy to build near-complete dossiers on unfamiliar animals within days. Structural scans from DeepPIV, behavioral footage from midwater cameras, and tissue-level images from instruments like Squid can all be assembled into a digital package that accompanies any physical specimen.
AI systems trained on vast image archives then provide a second layer of scrutiny. By comparing new observations against thousands of labeled examples, these models can quickly rule out common species and highlight edge cases that deserve expert attention. In principle, that triage should free taxonomists to focus on the most informative material and reduce the backlog of unprocessed cruise data that has historically slowed the description of deep-sea fauna.
However, the same tools that accelerate discovery also raise new standards for evidence. High-resolution imagery makes subtle differences in body plan and behavior more visible, which can tempt researchers to split lineages into new species on the basis of traits that may fall within natural variation. To avoid over-fragmentation, future expeditions will likely need clear protocols for when imaging alone is sufficient to propose a new species and when genetic or morphological confirmation from preserved material is required.
Designing the Future 3 underscores that tension. Its 31 candidate species are both a testament to technological progress and a reminder that formal taxonomy remains conservative by design. Until detailed descriptions, specimen access, and AI performance data are made public, the cruise’s most important contribution may be methodological: demonstrating that integrated imaging and machine learning can bring the hidden biodiversity of the South Atlantic’s midwaters into sharper, if still provisional, focus.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
