In the past week, social media was abuzz with discussions about a purported “virtual embodied fly” after several posts went viral on X. These posts, amplified by accounts dedicated to AI hype, sparked excitement and confusion among users who seemed unsure about the nature of what they were witnessing.

The intriguing videos originated from Eon Systems, a tech company based in San Francisco. The firm claims it is on a mission to achieve “digital human intelligence” and aims to digitally replicate a mouse brain within just two years—a goal that many would consider highly optimistic. Alexander Wissner-Gross, one of the company’s cofounders, initially released the video, describing it as the “world’s first embodiment of a whole-brain emulation that produces multiple behaviors,” suggesting a breakthrough toward technological singularity. Meanwhile, CEO Michael Andregg shared another version, calling it a “real uploaded animal.”

However, the claims were presented with only minimal evidence: short videos showcasing a digital fly seemingly engaging in behaviors like walking, feeding, and leg rubbing, but without any detailed methodology, scientific documentation, or independent validation.

In an attempt to clarify, Andregg posted a thread on X filled with qualifiers, ambiguous scientific jargon, and specific-sounding metrics like “91% behavior accuracy.” Despite these details, the meaning behind this figure remains elusive, even to those with a background in animal behavior, as I learned during my master’s studies. Nevertheless, Andregg reiterated, “this is, in our view, a real uploaded animal.”

The internet was abuzz. The evidence was still two short videos on X. If you’re going to tell the world you’ve just hit what would be one of the most significant scientific milestones in human history, you better come with receipts.

Andregg attempted to provide some clarity on X in a thread that was part grab bag of caveats, part vaguely-described scientific terms, and part concrete-sounding numbers like “91% behavior accuracy.” I’ve sat with that metric for a while, and I still don’t really know what it’s supposed to mean, and I spent a good chunk of my master’s degree studying animal behavior. Nevertheless, he insisted that “this is, in our view, a real uploaded animal.”

I sent Andregg a message on LinkedIn asking for more details. He replied with a link to a blog post Eon had just published titled “How the Eon Team Produced a Virtual Embodied Fly.” It wasn’t a scientific paper, but it was something, I guess.

For the experts The Verge spoke to, that blog wasn’t nearly enough, but it trod a much more cautious line than the posts on X, in so much as it doesn’t say “this is a real fly.” Shahab Bakhtiari, a professor leading the systems neuroscience and AI lab at the University of Montreal, said that while initial posts “obscured critical” details about the work, the new blog provided more context. “But it arrived a bit late, and remains insufficient to fully validate the claims,” he said. He would have expected a detailed technical report that includes details on things like software, code, and simulation environments that would let other scientists reproduce and evaluate the work.

Alexander Bates, a research fellow in neurobiology at Harvard Medical School studying fly brains, echoed Bakhtiari. He said the group “under-delivered” and that while the blog provided more details on what the team did — they stitched together existing large-scale projects like a detailed map of a fly’s brain, a physical simulation of a fly’s body, and models simulating how these interact in a virtual environment — “for a claim of this magnitude, I would expect something that should spell out the whole approach in specifics.”

Bates also said the virtual fly’s behavior should be evaluated against real data and “clearly defined metrics,” adding that the 91 percent figure was still not explained in the blog post. “Also, the fly does not fly.”

Bates told The Verge he understands that “strong framing and hype can matter for fundraising,” but stressed Eon’s claim of a “real uploaded animal” isn’t credible. Aran Nayebi, a professor of machine learning at Carnegie Mellon University, said that the group was “not even close” to capturing the full brain of the fly, showing connections between cells but not crucial details like neurotransmitters or how strong the connections between different nerve cells are. The motor system isn’t a “true upload” either, he said. “We are not even faithfully simulating its brain in silico.”

Fine, let’s say Eon actually did it. It copied the fly’s brain perfectly. The whole thing. Every last part. Do we have a digital fly now?

Yes. No. Maybe. I don’t know, and you probably don’t either. Neither does Eon. The blog conveniently glossed over the weighty definitional questions at the core of the upload claim: What, exactly, counts as a fly? When we think of flies, we don’t think of a bunch of behaviors or neural connections. We think of, well, a fly. Is it enough to reproduce a few fly-like behaviors in a simulation? Does a fully mapped brain in a virtual vat count? Or does “fly” mean the whole, messy biological package — a body, cells, metabolism, and whatever counts as “memory” or learned experience over the course of its life?

And that’s the easy version of the problem. The thing on the screen isn’t obviously a fly at all. It’s a composite of the neural wiring, programming, and other information stitched together from multiple different animals. This is useful when modeling, but precisely what organism can we meaningfully claim has been uploaded in this case? Technically speaking, it’s also a copy, not an upload, which comes with obvious and profound implications the hype conveniently skips over: You could make two, or 10, or 10,000 of the “same” fly. What then?

Normally, I wouldn’t expect a startup to solve a major metaphysical problem — philosophers have been squabbling over that one for centuries — but they’re the ones saying they’ve got a “real uploaded animal.”

The experts I spoke to weren’t even convinced the term makes sense. Bakhtiari said it’s still very much an “open question” as to whether a “real uploaded animal” is even possible. Jonathan Birch, a philosopher at the London School of Economics, was blunter: “I don’t think we should ever say ‘uploaded animal,’” he said. What Eon is aiming for, he said, is “whole-brain emulation,” leaving the rest of the animal behind.

That biology is important to behavior, said Tom McClelland, a philosopher at the University of Cambridge. “So at best they’ve uploaded some of the fly’s mind and thereby uploaded some of the fly.”

With some time behind his viral post, I asked Andregg if he stood by his claim. “Yes,” he told me. In fact, he went further: “We [the research group and its academic collaborators] think this fly is conscious in a limited sense, it can smell, see, taste, etc.” (I will absolutely not be going into the whole consciousness thing.) He described the system as a kind of “MVP,” minimum viable product, of an uploaded animal, one with “lots of limitations.” I don’t quite know how to picture a minimum viable fly. It’s a fly, not an app. MVP is tech startup speak, not science.

When I went back to Andregg a second time — this time after speaking with experts and relaying their criticism — he still stood by the original claim, but with even more caveats. He conceded the work “isn’t a perfect replica of a fly,” and added that Eon had never said it was. “I don’t think of uploading as a binary concept,” he told me, describing “different levels” of upload and admitting that we don’t yet know how much biology is required to capture the information that matters. “There is a lot more work to be done to achieve the level of upload that we may want for ourselves someday.”