Brussels Mandates AI Disclosure Labels to Combat Deepfakes and Misinformation

The August Mandate: Transparency and Penalties

Effective August 2, the European Union initiates a strict regulatory phase under Article 50 of the AI Act (AIA). The new Code of Practice on Transparency of AI-Generated Content forces technology developers to identify synthetic media, aiming to preserve the integrity of the digital information sphere. Failure to comply with these transparency obligations carries heavy financial risks, with administrative fines reaching as high as €15 million.

Defining Scope and Categorization

The European Commission identifies two primary pillars of concern: deepfakes and manipulated text. Deepfakes encompass AI-generated or altered audio, images, and video that convincingly mimic real individuals, locations, or incidents. Manipulated text refers to any AI-generated output released to the public without human oversight. To assist users, the EU has specified a trio of visual disclosures:

• Basic “AI” icon: Indicates the technology played a minor role in creation.
• “AI generated” icon: Reserved for products created entirely through prompts with no human involvement.
• “AI modified” icon: Used when human-made content has been altered via AI tools.

Exemptions for Art and Security

The code avoids a blanket ban on anonymous AI by providing specific legal pathways for creative and state functions. Satirical, fictional, and artistic works are granted simplified disclosure rules that ensure transparency without compromising the work's aesthetic display. Furthermore, the mandate does not apply to content authorized for criminal investigations, law enforcement, or the prevention of offenses. AI-generated text also escapes the labeling requirement if it remains under the editorial control and legal liability of a human reviewer.

Global Regulatory Shifts: EU vs. India

The European approach contrasts sharply with recent shifts in India. While India's Ministry of Electronics and Information Technology amended its IT Rules in February to address synthetic information, the focus remains on the platform and the user. The Indian model relies on social media firms to police and remove content, whereas the EU targets the AI developers directly. Furthermore, the EU's nuanced exemptions for artistic expression are absent from the more generalized Indian regulations.

Industry Integration and Resistance

Silicon Valley's response remains mixed. Google, while adhering to the code, warned through its EU head of Government Affairs, Karen Massin, that overlapping labels could confuse consumers and stifle European competitiveness. OpenAI expressed support but highlighted technical hurdles, noting that metadata can be stripped and digital icons tampered with. The organization emphasized that the code must remain flexible as the industry works to improve the reliability of provenance techniques.

Source: The Hindu — Sci-Tech