Onpode
Cover art for The iconic first black hole image just passed fresh scrutiny — here's what researchers confirmed

The iconic first black hole image just passed fresh scrutiny — here's what researchers confirmed

November 12, 2026 · 8 min

Cole Brennan & Malcolm Reeves

Hey, good to be back — how's your week been, honestly? Man, it's been a lot — but I'll tell you what, I've had this thing rattling around in my head since I stumbled onto it, and I genuinely cannot let it go. What thing? So someone showed a machine — an AI — photos…

On November 11, 2026, a study published in Nature by a small team of four researchers from Australia's national science agency, CSIRO, announced an independent confirmation of the iconic 2019 Event Horizon Telescope (EHT) image of black hole M87*.

0:008:16
Get the next episode on Science →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Science →

About this episode

On November 11, 2026, a study published in Nature by a small team of four researchers from Australia's national science agency, CSIRO, announced an independent confirmation of the iconic 2019 Event Horizon Telescope (EHT) image of black hole M87*.

Grounded in 10 sources
Nature publishes independent imaging algorithm paper confirming original EHT black hole image ↗ · nature.com
Iconic first image of black hole holds up to scrutiny | Nature ↗ · nature.com
Re-analysis of black hole data proves iconic ring shape really is there | New Scientist ↗ · newscientist.com
Is this iconic black hole image accurate? New study delivers an independent result ↗ · theconversation.com
New CSIRO technique confirms iconic black hole image - CSIRO ↗ · csiro.au
Astronomers Reveal First Image of the Black Hole at the Heart of Our Galaxy | Event Horizon Telescope ↗ · eventhorizontelescope.org
The Image of the M87 Black Hole Reconstructed with PRIMO - IOPscience ↗ · iopscience.iop.org
M87* black hole: Perth CSIRO team’s AI trained on cats and dogs confirms Event Horizon Telescope image | PerthNow ↗ · perthnow.com.au
Independent technique validates landmark M87* image ↗ · phys.org
Scientists check their work and confirm it with update on iconic black hole image ↗ · theregister.com
Read transcript

Malcolm Reeves: Hey, good to be back — how's your week been, honestly?

Cole Brennan: Man, it's been a lot — but I'll tell you what, I've had this thing rattling around in my head since I stumbled onto it, and I genuinely cannot let it go.

Malcolm Reeves: What thing?

Cole Brennan: So someone showed a machine — an AI — photos of cats and dogs. Everyday stuff, nothing astronomical. And then pointed it at the most famous black hole image in history. And it got basically the same answer the original team got. After years of people arguing about whether that image was even real.

Malcolm Reeves: Yes. And the name you want here — the person who made that choice — is Samuel Lai. Postdoctoral fellow at CSIRO, Australia's national science agency. He led a four-person team, and the choice they made before they ever touched the black hole data is, I think, the actual story.

Cole Brennan: Wait, the choice before the data?

Malcolm Reeves: The choice about what the algorithm was allowed to know. Their algorithm is called GenDIReCT, and they trained it on broad everyday images — cats, dogs, ordinary photographs — deliberately. No ring. No black hole. No prior expectation of what it should find. It's like handing someone a jigsaw puzzle with the box turned face-down. You want to see what they build without the picture on the lid.

Cole Brennan: So the whole point was to keep it naive. That's — yeah, that's the elegant part, isn't it. And they published this in Nature, November 11th, 2026, and it matched.

Malcolm Reeves: Matched — but here's what that matching was actually answering. Because the reason anyone needed GenDIReCT in the first place is that in 2022, Junichiro Makino at Kobe University did something uncomfortable. He took the original EHT data — the exact same dataset, April 10, 2019 — and ran it through one of the original EHT's own techniques. Just adjusted the assumptions slightly. And he didn't get a ring.

Cole Brennan: He got a blob.

Malcolm Reeves: A diffuse blob. No ring structure. And that result was published — and not retracted.

Cole Brennan: Wait, I want to actually sit in that moment for a second. Picture a grad student in Makino's lab, Kobe University, late night, 2022. They're running this reanalysis, right, probably expecting to just — confirm the ring. And instead this smeared blob comes up on the screen. I mean, what do you even do with that?

Malcolm Reeves: You publish it. Because the data hasn't changed. What's changed is the assumption baked into how you're asking the question. And that's precisely the problem — same ingredients, different recipe, different dish. The blob didn't go away. It sat there for years as a competing interpretation.

Cole Brennan: And nobody resolved it? Like, the EHT team didn't come back and say — no, here's why Makino's wrong?

Malcolm Reeves: Not in a way that closed it. The CSIRO paper says it 'largely refutes' the blob interpretation — that's the language they use. But as of publication, Makino's team hadn't publicly responded. So that door is still technically open.

Cole Brennan: That's — I mean, that's the part that shifts everything for me. Because if Makino's result is just sitting there unretracted, then what the EHT released on April 10, 2019 wasn't settled science. It was a strong answer to a question that was still, quietly, being contested.

Malcolm Reeves: And that's exactly why a truly independent method — one with no prior expectations about M87*, no ring-fitting bias baked in — that's why it matters. GenDIReCT wasn't just confirming the EHT's work. It was weighing in on a live dispute.

Cole Brennan: And that's the part — weighing in on a live dispute — that's where the naivety thing becomes way more than a design quirk. Like, GenDIReCT isn't just a clever method. It's almost like a character witness with no stake in the case.

Malcolm Reeves: And the number that lands it is 42.2 microarcseconds. Plus or minus 1.5. That's what Lai, Thyagarajan, Ivy Wong, Foivos Diakogiannis — that four-person team — that's what they pulled out of the original EHT dataset. Same angular diameter the EHT had. Black hole mass of 6.5 billion solar masses, plus or minus half a billion. The machine that learned from cats found the same number that hundreds of people with eight synchronized radio telescopes found.

Cole Brennan: Wait — how small is 42.2 microarcseconds actually?

Malcolm Reeves: It's the angular width of a human hair. Held at arm's length. From ten kilometers away. Across 55 million light-years.

Cole Brennan: I mean — that's insane. That's the number a naive algorithm landed on.

Malcolm Reeves: And Laurent Loinard — who praised GenDIReCT specifically — described this recovery as independent across several years of EHT data. Not one pass. Multiple years. The ring kept coming back.

Cole Brennan: And that connects back to Katie Bouman, right? Because she built the original multi-pipeline verification in 2019 — like, the whole standard of running multiple independent algorithms was hers. So GenDIReCT is almost — actually, wait — it's the logical next level of what she started. She said methodological independence is enough. GenDIReCT is the first thing genuinely orthogonal enough to test that.

Malcolm Reeves: Bouman established that methodological independence — not data independence — is the bar. And researchers called this 'the first truly independent validation' precisely on those terms. But there's a version of that claim we haven't fully pressure-tested yet — what 'independent' even means when everyone is pulling from the same underlying interferometric dataset.

Cole Brennan: Yeah — that's the thread I keep pulling on and not quite getting to.

Malcolm Reeves: And that's exactly the word that keeps slipping — 'independent.' Because the EHT built its virtual Earth-sized telescope using very-long-baseline interferometry. Eight radio observatories. That data got collected once. And every analysis since — Makino, the earlier ring recoveries before 2026, GenDIReCT — they're all working from that same publicly available interferometric dataset. Different pipeline. Same ingredients.

Cole Brennan: Three chefs, same kitchen. You haven't proven the food is good — you've proven the recipes can agree.

Malcolm Reeves: And Nature's own news coverage used that careful language — not dismissive, but cautious — about whether methodological convergence on existing data actually constitutes conclusive validation. The journal published Lai's result and still hedged the framing.

Cole Brennan: Wait — Nature published it and was still like, 'hold on though'?

Malcolm Reeves: Cautious. Not dismissive. And honestly, I think that's the honest position. Because the actual test — the one that settles it — is new observations. Fresh data. New conditions.

Cole Brennan: Which is — actually, that's the Sagittarius A* comparison. May 2022, EHT images a second black hole, the one at the center of our own galaxy. That's a genuinely new object. New data. That's not re-running the same archive — that's a second witness walking into the room.

Malcolm Reeves: Exactly the distinction. Sagittarius A* extended the portfolio — different object, different conditions, fresh observations. M87* still hasn't had that. What would it mean to do that for M87*? Upgraded arrays. New baselines. New collection. That's when you find out whether the ring was nature or our methods talking to each other.

Cole Brennan: And until then — I mean, the cumulative picture is real, the earlier ring recoveries, GenDIReCT, 42.2 microarcseconds coming back again and again — but we're essentially stacking methods, not experiments. That's a different kind of confidence.

Malcolm Reeves: It's the actual shape of twenty-first century science when the data costs too much to repeat quickly. And maybe — maybe that's the image that sticks. Not the black hole. The fact that four people in Perth and eight observatories spanning the planet all converged on the same number from the same archive. That convergence is the story. Whether it's enough is still an open question.

Cole Brennan: Four people. Eight observatories, hundreds of scientists, years of work — and four people in Perth with an algorithm that learned from cats and dogs got the same ring. 42.2 microarcseconds. That asymmetry is just... it's a lot to sit with.

Malcolm Reeves: And the ring is still not safe. That's the honest place this leaves us. The next real test isn't another method on the same 2019 archive — it's upgraded arrays, new baselines, genuinely fresh observations of M87*. When that data comes in, that's the moment. Either the ring is there, or it isn't. Either we were reading nature, or we were reading ourselves back to ourselves.

Cole Brennan: And someone showed a machine pictures of cats and dogs, and it found a glowing ring 55 million light-years away. I mean — whatever the future telescope finds — that detail is still just genuinely something. Thanks for working through this with me.