Deceptive Transparency: What the New AI Labels Achieve

• New EU AI Act regulations require mandatory disclosure for synthetically generated content, including text, images, video, and audio.
• AI providers and deployers must disclose when AI is used, such as in chatbots or biometric emotion/attribute recognition.
• Exclusions exist for purely personal use and minor AI-assisted editing tools where the original material is not substantially altered.
• Obvious non-realistic content, such as animations of talking animals, is exempt from labeling requirements.

Regulatory Scope

• Private individuals creating non-commercial content are not subject to the rules.
• Regulations apply in a tiered structure to "providers" of AI models and professional "deployers" using those systems.
• Guidelines clarify that misleading deepfakes of real people must be clearly labeled to prevent disinformation.

Effectiveness and Risks

• Experts warn that labels serve only as a starting point for evaluation, not a guarantee of truth, accuracy, or fairness.
• Standardized icons are recommended but not mandatory, leading to potential inconsistency and confusion across different platforms.
• Research suggests a "labeling paradox": content without labels may be perceived as more trustworthy, while true content with labels may be unfairly doubted.

Critical Perspectives

• Transparency is seen as a simple starting point rather than a complete solution, with researchers advocating for more robust system-design mandates.
• The AI Act is currently viewed as an experimental approach to integrating AI into society, with its real-world impact remaining to be seen.
• Media literacy is considered more essential than relying on labels alone to combat disinformation effectively.