This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Social media platforms increasingly rely on automated AI systems to remove terrorist and extremist content. However, these systems frequently misidentify human rights documentation and evidence of war crimes as prohibited content, leading to the permanent deletion of critical evidence. This lack of archival mechanisms hampers international criminal investigations and accountability efforts, as researchers and investigators lose access to irreplaceable documentation of atrocities.
Automated permanent removal of violating social media content, such as terrorism, violent extremism, and hate speech, without archival has allegedly hindered the potential use of this content for investigating serious crimes and hampered efforts in criminal accountability.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
Also classified as
Automated AI systems delete content instantly without human review, and users have no meaningful recourse or appeal to recover their lost evidence.
Inadequate regulatory frameworks and platform policies fail to mandate the archiving of deleted content for legal and accountability purposes.
The content moderation AI systems lack the capability and robustness to distinguish between prohibited extremist content and critical human rights documentation, leading to erroneous deletions.
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
The automated content moderation AI systems unintentionally deleted critical war crimes documentation while pursuing the goal of removing extremist content post-deployment.
Added categories peak at 3: Substantial (Psychological Harm, Inferred)
Automated AI content moderation systems on major social media platforms unintentionally deleted hundreds of thousands of videos documenting war crimes and human rights abuses. This irreversible loss of digital evidence severely hinders international criminal investigations, OSINT intelligence gathering, and individual asylum applications, presenting substantial long-term challenges to global human rights accountability.
The content moderation software operates autonomously to scan, flag, and delete uploaded videos in real-time based on learned patterns of graphic or extremist content, without requiring human intervention or planning for each action. As YouTube noted, '80% of videos flagged by its AI were deleted before anyone had seen them'.
High Risk: The AI systems are used in law enforcement and fundamental rights contexts, specifically content moderation affecting freedom of expression and the right to an effective remedy. The report describes how 'these platforms are taking down online content... in a way that prevents its potential use to investigate serious crimes, including war crimes', directly impacting fundamental rights and public interests.
The AI systems are explicitly deployed by social media platforms to moderate content, specifically to detect and remove terrorist, extremist, and hateful content.
The reports from the BBC, Reuters, and Human Rights Watch provide highly detailed, consistent, and well-documented accounts of the automated content moderation failures across multiple conflict zones. They include specific statistics, interviews with affected individuals, and statements from tech companies, leaving little ambiguity about the AI's role and the resulting harms.
None. The consensus across all three reports is clear that automated AI moderation is the primary driver of the erroneous deletions.
Technology
Context-Blind AI Algorithms
AI fails to distinguish war crimes documentation from extremist propaganda.
Broad Automated Filters
Algorithms block uploads or delete content before human review is possible.
Flawed Automated Appeals
Automated complaint systems reject appeals to restore mistakenly deleted video.
Data Inputs
Lack of Contextual Metadata
AI systems lack localized context data to accurately evaluate graphic uploads.
Human Factors
Over-reliance on AI Moderation
Platforms relied heavily on automation, reducing human content moderation.
Lack of Human Nuance
Machines lack human ability to identify human rights violations in video.
Process and Methods
Inadequate Archiving Systems
No mechanism exists to preserve AI-deleted content for legal investigations.
Opaque Takedown Procedures
Lack of transparency regarding how algorithms flag and remove war footage.
Ineffective Appeal Processes
Users face automated rejections when trying to restore deleted evidence.
Management
Prioritizing Speed Over Accuracy
Management prioritizes rapid automated removal over moderating accuracy.
Failure to Build Archives
Platforms fail to establish mechanisms to preserve evidence for investigators.
Regulatory Environment
Government Pressure for Speed
State demands to remove extremist content quickly force reliance on rapid AI.
Restrictive Data Retention Laws
Legislative restrictions limit how long platforms can store deleted content.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
