X's AI chatbot Grok generated a false headline claiming 'Iran Strikes Tel Aviv with Heavy Missiles' which was then promoted on X's main feed through the Explore feature on its first day of launch, spreading misinformation to millions of users.
On April 4, 2024, X (formerly Twitter) launched an updated version of its Explore feature that uses the AI chatbot Grok to generate headlines and summaries for trending topics. On the first day of the feature's launch, Grok created a false headline stating 'Iran Strikes Tel Aviv with Heavy Missiles' based on misinformation being spread by verified X users. The fake story was then promoted on X's main feed and homepage sidebar, where hundreds of millions of daily users could see it. The misinformation appeared to originate from blue checkmark accounts posting the same copy-and-paste false information about Iran attacking Israel, along with unverified videos. X's algorithms detected this as a trending topic and Grok automatically generated an official-looking headline and summary, which was then prominently displayed to users as if it were real news. The incident occurred in the context of real tensions between Iran and Israel following Israel's airstrike on Iran's embassy in Syria earlier that week, making the false claim seem plausible. X includes a disclaimer that 'Grok is an early feature and can make mistakes' and advises users to 'verify its outputs,' but the false headline was presented in a news-like format that could easily mislead users.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed