OpenAI to Implement ChatGPT Watermarks in EU, Global API Opt-In

Warsaw, Poland – September 20, 2024: A smartphone displaying OpenAI’s logo in front of the European Union flag, symbolizing innovation and technology in Europe. — Image credit: Search Engine Journal

BRUSSELS — OpenAI will implement invisible watermarks for texts generated by its ChatGPT and Codex models within the European Union in the coming weeks, responding to anticipated regulatory requirements.

The move comes ahead of new transparency rules under the EU AI Act, which are set to apply from August 2, 2026, and aligns with the voluntary Code of Practice on Transparency of AI-generated Content.

Globally, users of OpenAI’s application programming interface (API) can now opt-in to activate watermarking for select models, although this feature remains off by default, the company stated.

OpenAI’s text detection tool, named ‘textGrain’, will not be publicly available at its launch. Access will be restricted to approved researchers and expert organizations, according to the company.

Tests conducted by OpenAI revealed limitations in the detection tool’s effectiveness. Replacing 25 percent of words with synonyms in 400-token passages significantly reduced detection rates from 92 percent to 17 percent.

The European Commission has been developing comprehensive legislation for artificial intelligence, with other major AI developers like Anthropic and Google DeepMind also addressing the changing regulatory environment.

OpenAI’s proactive implementation in the EU is a direct response to Article 50 of the EU AI Act, which mandates transparency for AI-generated content, according to documents from the European Commission.

The company said in a statement that the EU-only rollout for its consumer-facing products is a strategic step to comply with these forthcoming regulations.

The watermarking system aims to help identify content generated by AI models, a growing concern amid the proliferation of advanced generative AI tools.

Source: Search Engine Journal