Researchers Confirm First Black Hole Image Using AI

An Australian-led team reexamined original Event Horizon Telescope data with a machine learning method and reproduced the M87* black hole structure.

Updated on Nov. 22, 2026 in Space

Photograph of white radio dish antennas tilted toward a star-filled night sky, with a faint glowing ring hovering in the distance above them.
Researchers led by Australia's CSIRO used an AI method to reanalyze original Event Horizon Telescope data, independently reproducing the 2019 image of the M87* black hole. AI Illustration. Upload story photo >

A research team led by Samuel Lai of CSIRO independently confirmed the first image of the M87* black hole by applying a machine learning method called GenDIReCT to the original Event Horizon Telescope dataset. The re-examined data produced a ring structure nearly identical in size and shape to the image released in 2019.

Why it matters

The result independently verifies the 2019 image using a method that had no prior knowledge of black hole imagery, strengthening confidence in the original observations. The reconfirmed structure also aligns closely with predictions from Einstein's general relativity.

The reanalysis compared a ring structure from the original dataset against the 2019 image, finding them nearly identical in size and shape, with M87* estimated at 6.5 billion solar masses. Whether the same method yields consistent results on other black holes remains untested in the reported findings.

The players

Samuel Lai

Samuel Lai leads the research team at Australia's CSIRO that conducted the independent reanalysis of the black hole data.

CSIRO

CSIRO is Australia's national science agency, which hosts the research team that confirmed the black hole image.

The details

The team developed GenDIReCT, a machine learning and artificial intelligence technique, to isolate structural information about the black hole while filtering distortions from Earth's atmosphere and biases from combining telescope data. The Event Horizon Telescope operates as a global network of radio telescopes functioning as a single Earth-sized instrument, and the findings were published Wednesday in the journal Nature.

Timeline

  1. 2015: Einstein's general relativity predictions described the structure of black hole shadows.

  2. 2019: The original Event Horizon Telescope black hole image was released to the public.

  3. November 18, 2026: The reconfirmation findings were published in Nature.

The Big Picture

The reanalysis reinforces the first Event Horizon Telescope image of M87*, released in 2019, by showing an independent method recovers the same ring structure. The agreement with Einstein's general relativity strengthens the case that the 2019 image reflected real black hole physics rather than instrumental artifacts.

The GenDIReCT method may eventually be adapted for medical imaging and for mapping seismic activity from earthquakes, though no timelines have been reported. For now, the findings mainly validate existing black hole observations rather than producing consumer-facing tools.

The takeaway

Independent reanalysis is a powerful way to test major scientific results, and this study shows a new machine learning approach can recover the same black hole structure from original data. Readers following space science can watch how such methods extend to other astronomical observations.

Further reading

Explore more space news and discoveries to see how black hole research fits into the broader field.

Source note: This article includes information reported by Nysun.