AI-Generated Images Available through Adobe Stock Misrepresent Real-World Events

AI-generated images available through Adobe Stock, depicting realistic but fictional scenes of real-world events like wars and protests, have raised significant ethical concerns. These images blurred the lines between reality and fiction in journalistic contexts, prompting Adobe Stock to "crack down on AI-generated images that seem to depict real, newsworthy events and take new steps to prevent its images from being used in misleading ways."

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.2 Pollution of information ecosystem and loss of consensus reality

The proliferation of thousands of unlabeled, photorealistic fake images of real wars and disasters across stock sites and social media threatens to pollute the information ecosystem and blur the lines between fiction and reality.

Additional risk subdomains

• 6.3 Economic and cultural devaluation of human effort: The potential replacement of professional photojournalists with cheap, synthetic AI-generated stock images devalues human journalistic effort and threatens the profession's economic viability.

Causal factors

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

The risk of widespread misinformation and erosion of public trust is an unexpected and unintentional outcome resulting from the post-deployment use of photorealistic image generation models.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves AI-generated content and deepfakes, which under the EU AI Act are subject to specific transparency and labeling obligations to ensure users are aware they are interacting with synthetic media.

AI system and alleged parties

• AI system: Dall-E, Stable Diffusion, unspecified (OpenAI, Stability AI)
• AI purpose: Deepfake Image Generation; Image Generation
• Behaviour type: Tool
• Alleged developer: Various AI image generators
• Alleged deployer: Adobe Stock
• Alleged harmed parties: News sources, Journalistic integrity, General public

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 Minor
• 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 Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the creation and spread of photorealistic fake images of war zones and disasters.

Directly caused: AI systems generated thousands of photorealistic fake images depicting graphic scenes of war, distressed children, and explosions in Gaza and Ukraine.

Indirectly caused: These fake images were hosted on commercial stock sites and subsequently shared on social media platforms like TikTok and Instagram.

Democracy

Reported: The report explicitly describes concerns regarding the erosion of public trust in journalism and the creation of false narratives.

Indirectly caused: The proliferation of unlabeled AI fakes on stock sites threatens to undermine the credibility of reputable news organizations and photojournalism, which are critical for an informed democratic public.

Inferred additional harm: Continued spread of realistic fakes could lead to a broader societal loss of shared consensus reality, weakening democratic discourse.

Epistemic

Reported: The report explicitly describes epistemic harm through the generation and spread of false photorealistic images depicting real-world conflicts.

Directly caused: AI models generated thousands of fake images of the Gaza conflict, Ukraine war, and Maui wildfires, presenting fictional scenes as real photojournalism.

Indirectly caused: These fake images misled social media users, including a prominent actress who reshared a fake Gaza image believing it was a real photograph.

Inferred additional harm: The widespread availability of photorealistic fakes erodes the general public's ability to trust any visual evidence online, leading to a broader epistemic crisis.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 1300000

Potential causes

Management

• Prioritizing Growth Over Safety: Embraced AI submissions for market growth before establishing safeguards.
• Reactive Risk Management: Policies were updated only after public backlash and media inquiries.

Technology

• Hyper-realistic Image Generation: Rapid advances in AI make fake images indistinguishable from real photos.
• Lack of Indelible Watermarks: Current AI tools lack robust, unremovable markers to identify fake content.
• Easy Metadata Removal: Downstream users can easily strip metadata labels from AI-generated images.

Data Inputs

• Mislabeled AI Submissions: Users uploaded AI-generated images without the required generative AI labels.
• Lack of Training Verification: AI models trained on stock libraries generate highly realistic fake events.

Human Factors

• User Sharing of Unverified Media: Social media users share AI fakes believing they are real photojournalism.
• Financial Temptation for Media: Budget cuts tempt cash-strapped media outlets to use cheap AI stock images.
• Cognitive Bias and Trust: People assume stock platforms only host authentic photojournalism.

Process and Methods

• Reliance on User Self-Labeling: Platforms relied on uploaders to honestly tag images as AI-generated.
• Inadequate Verification Filters: Stock sites lacked automated tools to detect and flag unlabeled AI images.
• Permissive Initial Content Policy: Policies allowed AI depictions of real-world sensitive, newsworthy events.

Regulatory Environment

• Fragmented Platform Policies: Inconsistent rules across platforms create gaps in AI content regulation.
• Absence of Industry Standards: No unified legal framework exists to mandate AI watermarking or labeling.

Information quality

• Classification confidence: High
• Reason for confidence: The report from The Washington Post provides clear, detailed, and well-sourced information about the presence of AI-generated images on stock sites, platform policies, and specific real-world examples of users being misled.
• Ambiguities identified: The exact number of news blogs that published the fake Gaza explosion image without labels is not fully specified.
• Alternative interpretations: None, the facts clearly point to a systemic issue of synthetic media polluting stock image databases.

The hosting of unlabeled, photorealistic AI-generated images of real-world conflicts on stock platforms like Adobe Stock led to their dissemination on news sites and social media. While it does not represent a direct physical or state-level threat, it highlights an evolving capability in information pollution that erodes public trust in media and consensus reality.

• Overall national security impact: Minor
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: Global
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The threat of AI-generated misinformation is an ongoing, long-term strategic concern for the information ecosystem rather than an immediate crisis.
• Autonomy: Human-controlled. The AI systems acted as tools directly controlled by human users who prompted the generation and uploaded the images.
• Novelty: Evolved capability. The use of generative AI represents an evolved capability of a long-standing threat of photographic manipulation and misinformation.

Impact by dimension

• Physical security: Negligible. No physical security threats, kinetic attacks, or critical infrastructure damage were reported in connection with these synthetic stock images.
• Information security: Minor. Unlabeled synthetic images of active conflicts were published on news blogs and social media, contributing to information pollution, though no coordinated state-sponsored campaign was identified.
• Sovereignty: Negligible. No direct threats to state authority, government decision-making, or electoral systems were reported.
• Economic security: Negligible. While the incident raises concerns about the livelihoods of photojournalists, it does not pose a threat to national economic stability or strategic industries.
• Societal stability: Minor. The proliferation of realistic fake images of conflicts risks eroding public trust in journalism and shared consensus reality, representing a minor threat to societal cohesion.