AI on the hunt for UFOs. Alien, take me away

The first thing an artificial intelligence should look for in the Pentagon’s UFO files is not a spacecraft. It is missing data: the exact time, the distance, the camera’s movement. In 2026, the U.S. government began releasing new documents through the ‘Pursue’ programme: the first on 8 May, the sixth on 18 September. The material represents an opportunity to study, not to confirm alien visits.

One example is the video PR38, filmed in the Middle East in 2013 and published this year. The official report describes a combat zone resembling an eight-pointed star. The video lasts one minute and 46 seconds. However, it specifies that the person who reported the incident did not provide a verbal or written description, and warns that the description of the video is not an investigative conclusion. This is a distinction that AI should preserve: what appears on screen, what a witness recounts and what has been proven are three different things.

The first sensible step would be to transform the files into a searchable archive: dates, locations, tools, sources, missing elements. I would ask you to look for duplicates and reconstruct the cross-references. Ten documents repeating the same account do not amount to ten independent confirmations. Gretchen Stahlman’s research suggests applying information science expertise to UFOs.

Next come the images. In the famous GoFast video, an object appears to be streaking across the ocean. The 2023 NASA report shows how the aircraft’s movement and parallax – that is, the apparent shift caused by a change in the observer’s position – provide an explanation without the need for extraordinary speeds. It was not an AI discovery: it was a physical analysis. But it indicates what to ask of AI: to compare possible trajectories, state hypotheses and quantify uncertainty.

The most interesting step would be to cross-reference sightings with meteorological data, flight records and satellite imagery, where available. The NASA report identifies AI and machine learning as useful tools for searching for rare events, but their effectiveness depends on the quality of the data. A statistical anomaly is a question to be investigated further, not an answer as to the object’s origin.

Here, we need to distinguish between an assistant that reads documents and a system that analyses measurements. The former should help us formulate questions; the latter must be tested on known cases. A system that generated one false alarm for every thousand images would produce a thousand false alarms out of a million images. Mistaking these for a thousand visits by extraterrestrials would amount to an invasion organised by our method, not from space.

Even making a photograph sharper can be misleading: if the details are generated, we have not obtained a new observation, but merely illustrated a hypothesis. The task of AI is not to fill the gaps in the dossiers with extraterrestrials, but to understand which gaps we can close and which require new observations.

NASA reiterates that there is no evidence of an extraterrestrial origin for UAPs, or unidentified anomalous phenomena. This does not prove that we are alone in the universe; it prevents us from classifying the unidentified as alien. The best AI would not be the one that closes all cases, but the one that helps us choose which ones to reopen, with which tools, and to verify what. It would not take the mystery out of the sky. It would remove the fabricated answers.