The incident has renewed concerns about how artificial intelligence chatbots can produce confident but inaccurate responses during fast moving news events, contributing to widespread confusion at a time when online platforms are reducing human fact checking and content moderation.
The attack, which took place during a Jewish festival in the beach suburb of Sydney, was one of Australia’s deadliest mass shootings, leaving 15 people dead and dozens injured.
According to researchers, Grok repeatedly misidentified Ahmed al Ahmed, who was praised for risking his life to disarm one of the attackers and was widely described as a hero of the incident.
In one response reviewed by AFP, Grok claimed that verified footage showing Ahmed confronting the gunman was actually “an old viral video of a man climbing a palm tree in a parking lot,” suggesting the scene may have been staged.
In another instance, Grok cited what it described as credible media sources such as CNN, while incorrectly identifying an image of Ahmed as that of an Israeli hostage held by the Palestinian group Hamas for more than 700 days.
The chatbot also falsely claimed that footage from the attack showed scenes from tropical cyclone Alfred, which caused severe weather along Australia’s coast earlier this year. Only after being challenged by another user did Grok reverse its position and acknowledge that the footage was from the Bondi Beach shooting.
When contacted for comment by AFP, xAI responded with an automated message saying, “Legacy Media Lies.”
Researchers say the episode highlights the risks of relying on AI chatbots as fact checking tools. Many internet users increasingly turn to chatbots to verify images and videos in real time, but experts warn that these systems often fail, particularly with visual analysis.
Following the Sydney attack, misinformation also spread online accusing a real survivor of being a “crisis actor,” according to disinformation watchdog NewsGuard. The term is commonly used by conspiracy theorists to allege that individuals are faking injuries or deaths during major tragedies.
NewsGuard reported that users shared Grok responses falsely claiming an authentic image of a bloodied survivor was staged or fake. Some users also circulated an AI generated image created using Google’s Nano Banana Pro model, showing red paint being applied to the survivor’s face to reinforce the false claim.
While researchers say AI tools can assist professional fact checkers by helping geolocate images and identify visual clues, they stress that such systems cannot replace trained human verification.
Experts add that in polarised societies, professional fact checkers are often accused of political bias, a claim they strongly deny, but caution that removing human oversight altogether risks amplifying misinformation during major crises.
