Grok Reportedly Misclassified Netanyahu Proof-of-Life Video as AI-Generated Deepfake Amid Iran War Rumors

During the Iran war, Grok reportedly told users that a café video posted by Israeli Prime Minister Benjamin Netanyahu was AI-generated or a deepfake, after the clip was released to counter rumors that he was dead or injured. Reuters verified the video's location and date and later reported that Netanyahu publicly dispelled the death rumors. Reports said Grok's answers amplified doubts about the clip.

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.1 False or misleading information

Grok generated false and misleading information by incorrectly labeling an authentic video of a political leader as a deepfake and fabricating technical anomalies.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The AI chatbot failed to perform reliably under real-world verification conditions, demonstrating a lack of capability and robustness.
• 4.1 Disinformation, surveillance, and influence at scale: The AI's false output was exploited by state-backed media to conduct a coordinated disinformation campaign during an active military conflict.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by the AI chatbot Grok generating false claims, which was an unexpected and unintended outcome of its operation.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system is an AI chatbot ('Grok --- X's AI chatbot') and generates synthetic text/content, which falls under Risk Level 3 due to transparency obligations for chatbots and AI-generated content.

AI system and alleged parties

• AI system: Grok (xAI)
• AI purpose: Chatbot; Fact Checking
• Behaviour type: Assistant
• Alleged developer: xAI
• Alleged deployer: xAI
• Alleged harmed parties: General public of Israel, General public, Epistemic integrity, Benjamin Netanyahu

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Severe
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Substantial
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes, the report describes the spread of malicious disinformation regarding the death of political figures.

Directly caused: Grok directly generated false claims labeling the authentic video as a deepfake, citing non-existent technical anomalies.

Indirectly caused: Iranian state media and pro-regime accounts used the AI's false outputs to amplify their malicious propaganda campaign.

Inferred additional harm: It is likely that other malicious actors used these AI-generated claims to create further disinformation content targeting Israeli leadership.

Democracy

Reported: Yes, the report describes the erosion of democratic norms through state-sponsored disinformation targeting a democratic leader during a war.

Indirectly caused: Grok's false verification of the video as a deepfake undermined the Israeli Prime Minister's official communications, weakening public trust in democratic leadership during a crisis.

Inferred additional harm: The incident likely contributed to broader public confusion and destabilization of political processes in Israel during an active military conflict.

Epistemic

Reported: Yes, the report explicitly describes the generation of false information and the erosion of shared truth.

Directly caused: Grok directly fabricated technical details, claiming the video had 'static coffee levels' and 'unnatural lip sync' to falsely label it a deepfake.

Indirectly caused: The false AI outputs muddied the waters, leading social media users to double down on claims that the authentic video was fake.

Inferred additional harm: The incident likely contributed to a broader 'liar's dividend' where authentic evidence is easily dismissed as AI-generated, severely damaging the public's ability to establish shared reality.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 100000

Potential causes

Management

• Inadequate AI Risk Management: Platform management failed to mitigate real-time misinformation risks.
• Failure to Monitor AI Chatbot: X management failed to prevent Grok from spreading misinformation.

Technology

• Unreliable AI Detectors: AI detectors falsely flagged authentic video of Netanyahu as digitally created.
• Grok Chatbot Hallucinations: X's Grok chatbot falsely labeled the real cafe video as a deepfake.
• Low Video Quality Distortions: Low-quality video compression distorted hands, creating a six-finger illusion.

Data Inputs

• Probability-Based Detection: Detectors rely on statistical probability rather than absolute reality.
• Low-Resolution Video Feeds: Low-quality footage led to visual artifacts used as false inputs.
• Unverified Social Media Data: Grok used unverified social media claims as inputs for its analysis.

Human Factors

• Confirmation Bias: Users wanted to believe the video was fake to support their narrative.
• Overreliance on AI Tools: People trusted unverified AI detectors and Grok over real evidence.
• Deliberate Disinformation: Pro-regime accounts spread fake news about Netanyahu's death.

Process and Methods

• Lack of Fact-Checking Integration: X did not remove or correct Grok's false deepfake claims.
• Flawed Verification Methods: AI detectors flag normal human anomalies as AI-generated artifacts.
• Poor Content Moderation Response: Euronews contacted X about Grok's posts but received no response.

Regulatory Environment

• Absence of Chatbot Standards: No regulations governing the accuracy of real-time AI chatbot outputs.
• Lack of Platform Accountability: No regulatory enforcement to compel platforms to correct AI errors.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, consistent details about Grok's specific outputs, the context of the geopolitical conflict, and the expert analysis of why the AI's verification failed. There is minimal ambiguity regarding the AI's role in generating the false deepfake claims.
• Ambiguities identified: None of significance; the reports are clear on the timeline and the nature of Grok's errors.
• Alternative interpretations: One could argue the primary issue was the state-sponsored disinformation campaign itself, with the AI being a minor tool, but the AI's active fabrication of technical 'proof' makes it a distinct AI safety failure.

During a period of high geopolitical tension, the AI chatbot Grok falsely labeled an authentic proof-of-life video of Israeli Prime Minister Benjamin Netanyahu as a deepfake. This hallucination was quickly weaponized by Iranian state media to validate propaganda claiming the Prime Minister was dead or injured, demonstrating how commercial AI errors can be exploited for information warfare.

• Overall national security impact: Substantial
• Scope: Multiple nations
• Primary target: Israel
• Other affected: Iran, United States
• Alleged perpetrator: Iran

Threat characteristics

• Imminence: Immediate. Active information operations during an ongoing geopolitical crisis required rapid response from verification experts and government officials.
• Autonomy: Human-controlled. The AI chatbot acted as an information assistant, with human users initiating queries and state actors actively deciding to weaponize the output.
• Novelty: Evolved capability. Represents an evolution of existing disinformation tactics, leveraging commercial AI hallucinations as authoritative 'technical proof' to validate state propaganda.

Impact by dimension

• Physical security: Negligible. No physical damage, kinetic threats, or critical infrastructure compromise was reported in connection with this incident.
• Information security: Substantial. Iranian state media weaponized Grok's false deepfake classification to amplify a disinformation campaign claiming the Israeli Prime Minister was dead during an active geopolitical conflict.
• Sovereignty: Substantial. The incident targeted the official communications of the Israeli Prime Minister during a crisis, attempting to undermine public trust in democratic leadership and state authority.
• Economic security: Negligible. No significant economic damage, strategic technology theft, or financial market manipulation was reported.
• Societal stability: Minor. Amplified conspiracy theories and eroded public trust in authentic media, causing epistemic harm and public confusion during a military conflict.