OpenAI signs EU transparency code to benchmark AI content provenance
OpenAI has announced its support for the European Commission's Code of Practice on Transparency of AI-Generated Content to assist with implementation of the EU AI Act. The company highlighted its ongoing adoption of C2PA provenance standards since 2024 and its multi-layered approach using both metadata and watermarking. This move encourages ecosystem-wide cooperation for checking media authenticity and verifying AI-generated content.
Key Takeaways
- Adopts the European Commission’s voluntary code to ensure machine-readable marking for all AI-generated images and video.
- Implements C2PA provenance standards and SynthID watermarking to protect metadata integrity against file transformations.
- Provides a public verification tool at openai.com/verify for third-party authentication of AI-generated media.
- Follows OpenAI's 2025 commitment as the first U.S. firm to sign the EU’s General-Purpose AI Code of Practice.
Why It Matters
This move signals the shift from voluntary safety pledges to technical compliance ahead of strict EU AI Act enforcement. By standardizing C2PA metadata and watermarking, OpenAI is establishing a technical baseline for the streaming and social media ecosystem to detect synthetic media. For streaming platforms and advertisers, this interoperability is critical for localized compliance and maintaining platform trust as generative video tools enter high-volume production. Watch for the August 2, 2026, deadline when AI Act transparency obligations become legally binding for all providers in the EU market.
Additional Context
The European Commission published the final Code of Practice on Transparency on June 10, 2026, targeting Article 50 of the EU AI Act, which requires clear labeling of deepfakes and AI-generated text. Per the European AI Office, signatories must implement machine-readable marking and imperceptible watermarking to assist both automated systems and end-users in recognizing synthetic content. This regulatory push arrives as platforms move toward technical enforcement; for instance, TikTok integrated C2PA in early 2025 to automatically label AI-generated videos, a move that resulted in the tagging of over 1.3 billion clips by mid-2026, according to recent technical updates. OpenAI’s compliance strategy is closely tied to its video ambitions. While the company’s Sora 2 video model, released in September 2025, included C2PA metadata by default, the industry has struggled with 'watermark removers' and metadata stripping. Per Reuters in May 2026, the emergence of steganographic watermarking has become the preferred industry counter-measure, as it hides data within pixels to survive re-encoding. The EU Code of Practice specifically addresses this by encouraging a multi-layered approach, acknowledging that no single metadata standard is currently foolproof against intentional manipulation. Beyond technical markers, the regulatory landscape is tightening around the 'deployers' of AI—the agencies and platforms using these tools for public communication. As reported by Slaughter and May in June 2026, providers of AI systems already on the market by August 2026 may receive a four-month grace period to comply with marking obligations. However, non-compliance for new systems carries severe penalties, with the EU AI Act allowing for administrative fines of up to 3% of a company’s global annual turnover for transparency violations.
Read full article at openai.com