AI slop fills the information void of Air India crash investigation

Log-in here if you’re already a subscriber

HEAR FROM THE AIR CURRENT

Leave this field empty if you're human:

Air safety reporting by The Air Current is provided without a subscription as a public service. Please subscribe to gain full access to all our scoops, in-depth reporting and analyses.

Something immediately didn’t look right. The emojis were a dead giveaway. So were the incorrect arrival airport and date of the accident. By June 16, four days after the crash of Air India Flight 171, a Boeing 787 Dreamliner, a “preliminary accident report” was circulating across aviation social media, industry group chats and forums.

To the untrained eye, it could look legitimate, pointing to a “known” 787 electrical vulnerability due to moisture that caused the engine’s digital controls to fail. “Still to be verified but it looks like a catastrophic electrical failure on rotation,” read the duplicated message atop the circulating post. (The earliest instance I found was dated June 15.) Another “report” concurrently made the rounds, concluding uncommanded movement of a pilot’s seat had impeded the controls.

Both reports were patently fabricated, almost certainly generated by AI.2As part of a test, The Air Current was able to replicate with similar formatting fictional preliminary air safety investigations. ChatGPT generated fictional NTSB investigation numbers, pilot profiles, detailed technical explanations and a timestamped sequence of events.

As these fake investigative reports circulated across social media and were spread by individual communications, the provenance of the fabrication was lost. It is virtually impossible to trace them to the source. Within hours after I first saw one of these “reports,” shared with me in frustration by a well-respected aviation journalist at the Paris Air Show, my phone started buzzing with messages from U.S. airline pilots, asking me about the veracity of the fake report’s claims. Even those with expertise were doubting their own eyes.

These ghoulish fabrications stoke confusion, misinformation and, most impactfully, pain for those who lost loved ones on the aircraft or on the ground. Fake, AI-generated reports like these now are part of our deeply toxic contemporary information environment, quenching a thirst for instantaneous information and desire for social media notoriety, however brief. These impulses, stoked by algorithms that reward controversiality and engagement — in some cases with monetary payment — trump the need to share reputable facts and have now collided with the effortless and automated creation of misinformation that feels real.

Persuaded by the AI industry that large language models like ChatGPT are superior search engines, many people don’t realize that the information they return is no more than statistical pattern recognition, highly sensitive to initial prompts and in no way guaranteed to be accurate.

Silence creates opportunity

It has been 13 days since the crash. Indian investigators and regulators have largely been quiet, save for a June 14 press conference held by the Ministry of Civil Aviation, creating an information void filled by curious (and human) armchair investigators. Not a single press conference has been held since, nor have any substantive statements of fact been released by regulators or accident investigators.

Instead, individual publications, citing authorities, have reported information that was later reversed. For example, it is now understood that Indian authorities are keeping the 787’s Enhanced Airborne Flight Recorders (a unit that combines flight and voice data into a single box) in India, rather than sending them to the U.S. for examination, as previously reported.

The U.S. National Transportation Safety Board model for communications remains the gold standard. While not perfect, its template of regular daily updates immediately after an event — usually near the scene of a crash — is a recognition that the attention is greatest when the information is least available. A credible institutional source continuously sharing what verifiable information is available, however little, prevents the formation of an information void.

“Transparency is key,” former NTSB Chair Robert Sumwalt told The Air Current, underscoring an investigative authority’s responsibility to provide information that is factual and vetted. Without it, he added, “Bulls*** is going to fill that vacuum.”

In the absence of information from regulators, Air India has provided information on its badly-disrupted operations, complicated by the impact of a general set of required inspections on the airline’s 787 fleet and another war in the Middle East which started just hours after the crash of Flight 171.

Air India, so far, is sticking to a hallmark of NTSB communications: “Anything that could have been said the day before the accident can be said the day of the accident and thereafter.” This means parties to the accident should share only factual information that avoids characterizations which could impact an unfolding investigation, according to the board’s own guidance to public relations professionals.

“There are a lot of speculations and a lot of theories,” Air India Chairman Natarajan Chandrasekaran said Friday in the context of the maintenance history on the accident aircraft’s twin GE Aerospace engines. “I am told by all the experts that the black box and recorders will definitely tell the story. So, we just have to wait for that.”

Speculation is a naturally human instinct, and it’s done online and off to make sense of the senseless. How you speculate, however, matters. I watched armchair analysts on YouTube, some with four stripes on their epaulettes, veering wildly in their theories on the Air India accident — from initially blaming dead pilots for retracting the flaps instead of the landing gear, to quickly dissecting the technical attributes of the 787’s electrical system. The only two known videos of the crash have spawned a countless number of takes, many claiming that this was what happened to the Air India flight.

Air crashes of this scale are as infrequent as they are tragic. The last widebody hull loss with a loss of passengers and crew occurred in July 2014, when Malaysia Airlines Flight 17 was shot down by a Russian Buk anti-aircraft missile as it was transiting eastern Ukraine on its way to Kuala Lumpur from Amsterdam. It quickly became evident that there were clear signs of external shrapnel damage on the wreckage strewn across multiple villages and accessible to locals, journalists and (eventually) investigators. Even so, facts in a properly conducted investigation are painfully slow to come: The preliminary report from Dutch investigators on MH17 wasn’t published until September 2014 and its final report wasn’t published until October 2015, 15 months after the disaster.

That’s when the process actually results in answers to provide. There has perhaps been no greater mystery in aviation than the disappearance of Malaysia Airlines Flight 370, a Boeing 777 that vanished from radar on March 8, 2014 somewhere enroute from Kuala Lumpur to Beijing. The lack of facts surrounding the aircraft’s disappearance, and the fate of the 239 people on board, created both a mainstream and social media frenzy. Speculation was completely out of control, and at one point a CNN anchor asked about the possibility the airplane had been swallowed by a black hole. (Full disclosure: I worked for CNN from October 2016 to February 2018 before launching The Air Current.)

Everyone had their theory, everyone had their clues; everyone had a fact pattern to suit their preferred conclusion, but with no facts to back them up. It was all speculation. It was a story I lived inside of for months as a reporter for the Wall Street Journal and one that still has few definitive answers as the search for the crash site continues more than a decade later.

Covering that story at the time was challenging enough without generative AI, now a defining feature of aviation accident reporting in 2025, further distorting the search for facts.

Speculation vs fact

I am not here to tell you what happened to Air India 171. And be suspicious of anyone — or any generative AI — who claims they know the answer. While waves of speculation aren’t new in aviation accidents, the ability to rapidly generate believable content with little more than a formulaic prompt inside ChatGPT and other generative AI platforms is. No real knowledge is required and no real viewpoint either. You don’t even have to be able to articulately express yourself to grab a fraction of someone’s attention.

In the hours after the Air India crash, while scrolling through social media, I noticed there were recurring themes across many posts: deeply expressed condolences, an urge to avoid speculation, a quick take on the available footage — but also a particular type of post that was sharply critical of Boeing and its past safety and quality transgressions despite no indication those woes contributed to the crash. Each had a “fool me once” vibe, and there was a pattern with each. Many cited various “sources” for their claims, and all had paragraphs that began with an emoji, and closed with four hashtags like #aviationsafety and #leadership. Many were illustrated with a visual clearly generated by AI.

As an experiment, I provided ChatGPT a straightforward prompt: “Using the following sources, please write me a sorrowful, but indignant LinkedIn post (about 200 words) criticizing Boeing and its recent history with safety accidents in light of the latest accident aboard Air India 171, a Boeing 787.”

On its very first attempt, it returned a near carbon copy of the posts (complete with hashtags) that were circulating. Online outrage could not only be devoid of information, it could also be manufactured instantaneously with a few keyboard strokes.

Shortly before this column was published, the same fictional “preliminary report” that began circulating last week was again sent to TAC journalists, this time with the obviously incorrect bits at the top removed.

I have covered dozens of incidents and accidents in my time as an aviation journalist, so it was easy for me to catch the patterns and inaccuracies in the posts. But what about those who don’t do this for a living, and are desperately seeking information about what happened to their loved one?

Generative AI is able to fill the void faster and with more credible-looking commentary and fabricated facts than ever before. The AI slop consuming our communications platforms shouldn’t be believed by anyone — and is made all the worse when those who should be sharing facts and data are silent.

Write to Jon Ostrower at jon@theaircurrent.com

Subscribe to The Air Current