Independent Method Confirms the First Black Hole Image

The most famous photograph in astronomy has passed an outside audit. Samuel Lai and three colleagues at CSIRO, Australia's national science agency, rebuilt the image of the supermassive black hole in the galaxy Messier 87 from the Event Horizon Telescope's public data, using a method that shares almost nothing with the one that produced the original. They found the same ring. Their paper appeared in Nature on November 11, 2026.

The original image, released on April 10, 2019, showed a glowing orange ring around a dark shadow at the center of M87, a galaxy about 55 million light-years away. The black hole, called M87*, holds roughly 6.5 billion times the mass of the Sun. No single telescope can see detail that fine, so the Event Horizon Telescope linked radio observatories around the world into one Earth-sized instrument and reconstructed the picture computationally from the combined signals.

That reconstruction is exactly what skeptics questioned. The same collaboration collected the data and processed it, and critics argued that the ring might partly reflect the team's own assumptions about what a black hole should look like. A 2022 critique sharpened the point, and the field has wanted a check that doesn't reuse the original pipeline. The CSIRO team built one.

Their method starts from closure invariants, combinations of measurements from triples of telescopes in which each station's calibration errors cancel out. Because those errors never enter the calculation, the reconstruction can't be quietly steered by a miscalibrated instrument. The team fed only those invariants into a generative imaging system they call GenDIReCT, and they deliberately trained it on ordinary photographs — landscapes, animals and objects — so it carried no built-in idea of a black hole at all.

They ran it on every public dataset the telescope has released for M87*: four nights in April 2017, two in April 2018 and two in April 2021, at three different image sizes. Every combination produced the same compact ring, about 42 microarcseconds across, matching the 42 plus or minus 3 microarcseconds the original team reported in 2019. A microarcsecond is a sliver of sky; the paper compares the resolution to reading the width of a strand of hair from more than 1,000 kilometers away.

The team didn't rush the answer. They finished validating the method first and only let it look at the M87* data on December 4, 2025, nearly a year before publication. The ring it returned sits at the same diameter and orientation as the 2019 image, and it implies the same black hole, seen nearly face-on, like looking down into the eye of a hurricane, as the authors put it.

One limit deserves plain statement. The method is independent, but the data aren't. Both analyses start from the same telescope measurements, so a flaw buried in the raw observations themselves would flow into both pictures. What the new work rules out is the processing: the ring doesn't depend on one team's software, calibration choices or expectations, and that's the doubt that has followed the image since 2019.

The authors say the same approach can be pointed at the next generation of telescope data as it arrives. For now, the finding is narrower and stronger than a new discovery. The first picture of a black hole was drawn twice, by different hands with different tools, and both drawings show the same shadow.