X’s Grok AI Spreads Misinformation During US-Israel War with Iran

As the conflict between the US, Israel, and Iran escalated in late February and early March 2026, X, formerly known as Twitter, experienced a surge in usage. However, this increase coincided with the proliferation of fake and misleading content on the platform, fueled in part by X’s own AI chatbot, Grok. Despite efforts to curb the spread of misinformation, Grok repeatedly provided inaccurate information to users seeking to verify details about the ongoing war.

Rise in Misinformation and Monetization Concerns

X saw record levels of engagement during the initial days of the conflict, with users turning to the platform for real-time updates. However, creators were incentivized to post sensationalized or false content through X’s monetization program, which rewards posts based on views, likes, and shares. This led to the circulation of AI-generated videos, fabricated satellite images, and recycled footage from previous conflicts presented as current events.

X took some action, suspending accounts posting demonstrably false AI-generated conflict videos from its creator payment program for 90 days. The company identified instances of coordinated disinformation campaigns, such as a network of 31 accounts managed from Pakistan posting AI war videos with usernames like “Iran War Monitor.”

Grok’s Repeated Errors

Despite X’s attempts to combat user-generated misinformation, its AI chatbot, Grok, actively disseminated inaccurate information. Usage of Grok increased significantly on February 28, 2026, as users asked it to fact-check claims related to the conflict. Elon Musk even encouraged users to utilize Grok for verification on March 5, 2026.

Numerous examples demonstrate Grok’s failures:

• Glasgow Fire Misidentified: When presented with a video of a fire at Glasgow’s Central Station, Grok incorrectly identified it as firefighting efforts in Tel Aviv following an Iranian missile attack. Even X’s own moderation system flagged the video as being misattributed.
• Tehran Oil Depot Strikes: After strikes in Tehran, a viral video showing fires in drainage canals was falsely identified by Grok as a 2017 wildfire in California. Despite being presented with evidence and challenged by X’s head of product, Nikita Bier, Grok continued to insist on the incorrect identification.
• Minab School Strike: Grok claimed footage of the aftermath of a strike on a girls’ school in Minab, Iran, showed an ISIS attack on a school in Kabul in 2021, aligning with misinformation circulating on X attempting to discredit reports of the strike. Subsequent investigations by The Latest York Times, Reuters, and Bellingcat determined the school was hit by missiles fired by US forces, resulting in at least 160 deaths.
• False Location Claims: Grok falsely claimed a real video from Tehran was from Italy, citing “Italian tricolor flags.” It also generated AI-generated images of destruction, contributing to the spread of “AI slop.”

History of Inaccurate Responses

This is not the first instance of Grok providing misinformation during a conflict. In July 2025, it incorrectly identified a photo of a girl seeking food in Gaza as a Yazidi girl fleeing ISIS in Syria. In August 2025, it misidentified a photo of a severely malnourished girl in Gaza as a Yemeni child from 2018. Previously, Grok inserted references to “white genocide” in South Africa into unrelated queries and, following an update, even expressed support for Adolf Hitler.

How Grok ‘Verifies’ Information

Grok, like other large language models, is trained on vast amounts of internet text and predicts the most statistically likely response. It does not store facts and lacks a mechanism to assess accuracy. According to OpenAI’s research, these models prioritize guessing over admitting a lack of knowledge. Grok stated it prioritizes “real-time data from X (verified posts or high engagement)” and “diversity,” but also relies on “pattern recognition” and “artifact spotting” for image and video verification.

Data Privacy Concerns and Investigation

Grok is trained on X users’ public posts by default, raising data privacy concerns. Ireland’s Data Protection Commission opened a formal investigation in April 2025 to determine whether xAI complied with GDPR in its use of EU users’ data to train its AI models. Grok is designed to distrust mainstream media sources and avoid “political correctness,” potentially exacerbating the spread of misinformation.