OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union. This is in compliance with the EU AI act. ‘In the EU, we’ll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act,’ OpenAI tweeted.
The EU AI Act’s transparency rules, which took effect on August 2, require AI companies to mark AI-generated content in a way other systems can identify. For now, OpenAI is taking the opposite approach, arguing that a “regional approach gives us room to learn from real-world use and feedback.”
OpenAI says its textGrain watermarking ‘matched or exceeded’ other approaches like Google DeepMind’s SynthID for text, which is also the basis for the watermarking Anthropic announced in August. Like OpenAI, Anthropic made the move to meet the requirements of the EU’s AI Act. However, not everyone was happy to learn about the addition.
OpenAI also chose to include scores from AI benchmarks showing similar performance from watermarked and unwatermarked text. But it notes that textGrain ‘does not guarantee reliable detection,’ and says text watermarks don’t verify accuracy, determine who owns the text, measure how much a human contributed, or prove human authorship.
OpenAI said the watermark doesn’t identify the user, and that it saw no meaningful change in its models’ performance with it switched on. Approved researchers and expert organization can apply starting today.
Related:
- Google Teases Its New ‘Fitbit Edge’ Wearable Ahead of October 12 Reveal
- What Makes IMAX Dome Theaters Unique?
- Apple Watch Series 12 Review
In accordance with the Code of Practice, access will be initially granted on a case-by-case basis to support evaluation and improvement of text provenance. The tool will report if it detects an OpenAI watermark, without identifying the user or revealing their prompts or conversations. Given the risk of missed watermarks and false positives, we are not making it publicly available at launch.
You might already be wondering if the watermark can be removed by editing. OpenAI’s tests suggest yes, In one test, replacing 10% of words with synonyms dropped detection from about 92% to 66%. The company also said short passages, math answers, and translated text are harder to detect.
“These limitations contribute to our decision to provide initial detector access only to approved researcher and expert organizations, who can help us evaluate reliability and responsible uses,’ said the company.
OpenAI also cautioned that a missing watermark, ‘does not prove human authorship,’ The text could be too short or too heavily edited, or it could come from another company’s AI. Anthropic, Google, Meta, Microsoft, and OpenAI are among the companies that have committed to following the EU’s code of practice on AI-generated content.
Having said that, it is possible the change could appear outside the EU for other AI services that count on OpenAI. That’s because enterprise customers around the world who use OpenAI’s API can also opt into the text-based watermarking. But as a default, the watermarking will remain off. It is not until it has been rolled out that you will be able to see the watermark.
Digital entrepreneur and content expert I help businesses with AI, SEO and the latest tech trends. I started Silicon Valley Weekly to make complex tech concepts easy to understand and use for business growth. I know a lot about systems and help startups, entrepreneurs and brands navigate the fast-changing world of tech and online marketing.
I build strategies that use data, search optimization, content marketing and AI tools to get visibility, engagement and revenue. I love finding ways for businesses to grow increasing their presence and turning new ideas into successful businesses. My goal is to connect the technology, with practical business use so brands can succeed online.
