<>
Starting August 2, 2024, organizations operating in the European Union must comply with new transparency obligations under the EU AI Act. Providers and deployers of AI systems are now required to explicitly disclose when users are interacting with AI, or when they are exposed to AI-generated content like deepfakes, emotion recognition, or biometric categorization. Failure to meet these requirements can result in significant financial penalties, reaching up to €15 million or 3% of a company’s total worldwide annual revenue.
Transparency Obligations for AI Providers and Deployers
The EU AI Act mandates that any natural or legal person—including public authorities and agencies—that develops an AI system or places one on the market under their own name must ensure transparency. This applies to companies regardless of their geographic location, provided their systems are used within the European market. According to the European Commission, these measures aim to build public trust and reduce the risk of deception by ensuring citizens know when they are interacting with an AI agent or chatbot.
Systems covered by these rules include conversational agents, chatbots, and AI companions. However, the regulations provide specific exemptions for tools such as spam filters, search and retrieval functions, predictive maintenance, and authentication systems. Furthermore, while deepfakes must be labeled, the Act includes exceptions for content that is clearly “artistic, creative, satirical, or fictional.”
Labeling Requirements and Machine-Readable Standards
To ensure consistent identification of synthetic media, the Commission has introduced three specific labels: “AI,” “Fully AI-generated,” and “Partially AI-modified.” These labels are intended to clarify the level of human oversight involved in the creation of text, audio, images, and video. “Fully AI-generated” applies to content created without human intervention beyond the initial prompt, while “partially AI-modified” covers content where human work is altered by AI, such as deepfake face-swapping.
The Commission has made these icons available for free in PNG and SVG formats. Beyond visual labels, systems must incorporate machine-readable markers to ensure the provenance of the content remains detectable. While the core transparency requirements take effect on August 2, 2024, systems already on the market before this date have a grace period until December 2, 2024, to achieve full compliance.
Compliance Through Codes of Practice
The European Commission has introduced a voluntary code of practice to assist companies in demonstrating compliance. Signing this code provides organizations with a structured framework for labeling and marking practices, as well as access to a ‘Signatory Taskforce’ for industry collaboration. Organizations that choose not to sign the code must still prove that their internal methods for transparency are “adequate” when reviewed by surveillance authorities.
This involves creating a comprehensive inventory of all systems that interact with users or generate content, defining clear roles for providers and deployers, and ensuring that disclosures are presented at the first point of interaction. Because enforcement will vary by member state, companies are advised to establish a common baseline for transparency that can be adjusted with local jurisdictional overlays.
Key Takeaways for Enterprises
- Scope: The rules apply to any entity placing AI systems on the EU market, regardless of where the company is headquartered.
- Deadlines: Most transparency requirements go into effect on August 2, 2024; legacy systems have until December 2, 2024.
- Exemptions: Content involving human editorial control or specific artistic/satirical purposes may be exempt from certain labeling requirements.
- Documentation: Companies should maintain a central record of systems, evidence of human review, and testing results to satisfy potential audits by surveillance authorities.
As the deadline approaches, organizations are urged to treat the August 2 date as the primary target for compliance. Effective transparency requires more than just a label; it necessitates a verifiable chain of evidence that accounts for how content is generated, edited, and published across different platforms.
Worth a look