This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
The use of AI-enhanced and AI-generated imagery in political discourse regarding Iranian human rights abuses has created a 'liar's dividend' effect. By mixing authentic photographs of political prisoners with AI-enhanced or entirely synthetic imagery, political actors have inadvertently provided the Iranian judiciary with a pretext to dismiss all human rights documentation as fabricated. This degradation of the evidentiary record obscures the reality of ongoing state violence and harms the credibility of legitimate human rights advocacy.
A purportedly AI-edited collage of eight Iranian women political prisoners was reportedly amplified by President Donald Trump and U.S. government accounts alongside an unverified claim that all faced imminent execution. Reporting said the images were of real women but had been AI-enhanced, while later Iranian official mockery used AI-generated women. The episode reportedly blurred authentic human-rights documentation with synthetic material and helped Iran dismiss underlying abuses.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The mixing of authentic and synthetic content erodes the credibility of real human rights documentation, leading to a 'liar's dividend' where audiences withdraw in skepticism.
The incident involves coordinated and at-scale political influence and propaganda by state and non-state actors using AI-altered and AI-generated content to manipulate public opinion regarding human rights.
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 risk was caused by human actors intentionally using AI tools to enhance or generate images for political propaganda and trolling post-deployment.
Added categories peak at 4: Severe (Epistemic Harm, Indirect)
The incident demonstrates the 'liar's dividend' in a geopolitical context, where the mixing of authentic and synthetic AI-generated content allowed the Iranian regime to dismiss genuine documentation of state violence. This erodes the credibility of international human rights advocacy and misleads democratic diplomatic institutions, representing a substantial information warfare and societal stability threat.
The AI systems (ChatGPT and image editors) were used as tools by human actors to perform specific, well-defined functions of image generation and retouching based on direct inputs.
Risk Level 3: Limited Risk. The incident involves AI-generated content and deepfakes, which are subject to transparency and labeling obligations under the EU AI Act.
The AI systems were used to generate entirely synthetic images of women (ChatGPT) and to retouch/enhance real photographs of political prisoners.
The report provides highly detailed, verified journalistic analysis of the specific posts, the political context, and the names of the affected individuals, allowing for precise classification.
The incident could be viewed purely as a political disinformation campaign, but the central role of AI-generated and AI-enhanced content in creating the 'liar's dividend' makes it an AI safety issue.
Technology
Ubiquitous AI Photo Editing
AI retouching built into standard software leaves traces on authentic photos.
Generative AI Propaganda Tools
AI models allow rapid generation of synthetic videos and images for propaganda.
Inadequate Platform Detection
Social media platforms lack effective tools to detect and label AI content.
Data Inputs
Lack of Provenance Metadata
Images lack built-in cryptographic signals showing their authentic origin.
AI-Enhanced Visual Artifacts
Smoothing and stylized backgrounds mimic fully synthetic AI-generated images.
Unverified Social Media Feeds
Plentiful unverified digital media feeds are ingested without verification.
Human Factors
Confirmation Bias in Sharing
Political figures share unverified AI-enhanced content to fit narratives.
Erosion of Public Trust
Audiences withdraw and dismiss real human rights abuses as AI fabrications.
Lack of AI Literacy
Users cannot distinguish between AI-enhanced and fully AI-generated images.
Process and Methods
Deficient Verification Work
No human rights or journalistic verification was done before amplification.
Inconsistent Platform Labeling
Meta and X fail to consistently label AI-generated content during crises.
Weak Content Moderation
Platforms lack resources to separate synthetic from authentic media in crises.
Management
Prioritizing Speed Over Accuracy
US government feeds amplified posts rapidly without checking AI authenticity.
Underinvestment in Fact-Checking
Platforms and organizations underfund crisis-specific AI verification.
Slow Policy Implementation
Platforms fail to translate Oversight Board AI recommendations into action.
Regulatory Environment
Absence of Provenance Standards
Lack of mandated standards for content provenance at the point of creation.
Lack of Platform Accountability
No regulatory enforcement of platform policies regarding AI-generated media.
Weak Crisis Response Policies
Regulators fail to mandate fast guardrails during geopolitical crises.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
