This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
During the January 2025 Southern California wildfires, users on social media platforms circulated AI-generated images depicting the iconic Hollywood Sign on fire. These images, some of which were identified as being created by X's Grok chatbot, caused unnecessary panic and confusion among residents who were already facing evacuation orders and life-threatening conditions. Fact-checkers and local officials confirmed the landmark was never in danger, highlighting the role of generative AI in amplifying misinformation during public emergencies.
During the January 2025 Southern California wildfires, AI-generated images of the iconic Hollywood Sign being on fire circulated on social media.
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
The incident involved the generation and viral spread of false AI-generated images depicting a landmark on fire, misleading the public during an active emergency.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by human users intentionally generating and sharing fake disaster images using a deployed AI system.
Added categories peak at 2: Minor (Epistemic Harm, Direct)
Users leveraged X's Grok AI to generate realistic fake images of the Hollywood Sign on fire during the January 2025 Southern California wildfires. The viral spread of these images caused temporary panic and confusion among evacuating residents and required local officials to issue debunks, illustrating the growing risk of AI-amplified misinformation during public emergencies.
The AI system, Grok, was used as an image generation tool that responded directly to user prompts to create synthetic images, without exhibiting autonomous planning or agency.
Risk Level 3. Limited Risk: The report describes AI-generated content such as deepfakes or synthetic images of a disaster, which require transparency obligations so users know they are interacting with AI-generated media.
The AI system (Grok) was used by social media users to generate fabricated images of the Hollywood sign on fire based on text prompts.
The reports provide clear, consistent details about the generation of the fake images, the specific AI tool involved (Grok), and the public reaction. The facts are well-verified by fact-checkers and local authorities.
None
Technology
Grok Image Generation
Grok allowed users to generate realistic images of a landmark on fire.
Removable AI Watermarks
Grok watermarks were easily cropped or missing in viral versions.
Algorithmic Amplification
Social media algorithms spread the fake burning sign images rapidly.
Data Inputs
Unrestricted Prompts
Grok accepted prompts to generate disaster imagery during active fires.
Human Factors
Malicious Prompting
Users intentionally commanded the AI to create deceptive disaster photos.
Uncritical Sharing
Social media users reposted fake images without verifying authenticity.
Truth Apathy
Some users reposted fake images while expressing indifference to reality.
Process and Methods
Delayed Moderation
Platforms added fact-check disclaimers only after images went viral.
Reduced Fact-Checking
Meta's pivot away from fact-checking slowed down the verification process.
Management
Permissive Release Policy
Management released image generation tools without sufficient guardrails.
Safety De-prioritization
Decisions to pivot away from active fact-checking prioritized other goals.
Regulatory Environment
Lack of AI Standards
No active regulatory standards prevented the generation of disaster fakes.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
