June 26, 2025: Grok fact-check failures during the Israel-Iran conflict have sparked concerns over the chatbot’s reliability in verifying wartime information. A recent report by the Digital Forensic Research Lab (DFRLab) of the Atlantic Council exposed inconsistencies and factual errors in Grok’s responses to user prompts.
The study examined over 130,000 posts on platform X, where Grok fact-check is integrated. It revealed that the AI system frequently misjudged AI-generated visuals and spread unverified narratives. This included a video of a supposedly bombed airport that Grok alternately described as damaged, undamaged, or located in various unrelated cities such as Beirut and Tehran.
In one striking case, Grok even cited Yemeni missile attacks as the cause of the damage. However, in other replies, it failed to acknowledge any incident altogether. Such contradictions occurred within minutes of each other.
Another viral video — digitally created to show buildings collapsing in Tel Aviv — was labeled “authentic” by Grok, according to DFRLab. These incidents underline Grok’s poor handling of deepfakes and conflict-related misinformation.
False claims also spread that China had deployed military planes to Iran. When prompted, Grok and another AI system, Perplexity, wrongly confirmed the reports. This was flagged by disinformation monitor NewsGuard.
Previously, Grok made similar errors regarding protests in Los Angeles and tensions between India and Pakistan. Moreover, the chatbot once injected the term “white genocide” into an unrelated South African query, later blamed on unauthorized changes.
Elon Musk himself criticized Grok for citing Media Matters, an outlet he has sued. “Shame on you, Grok,” Musk posted.
