OpenAI is adding text watermarking in ChatGPT and Codex
OpenAI is adding an invisible, machine-readable watermark to text output in ChatGPT and Codex, according to The Verge. The rollout is limited to users in the European Union at first.
OpenAI says its textGrain watermarking "matched or exceeded" other approaches, including Google DeepMind's SynthID for text, which The Verge reports 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, the report says. Not everyone was happy to learn about the addition, according to The Verge.
OpenAI also included scores from AI benchmarks showing similar performance from watermarked and unwatermarked text. It notes, however, that textGrain "does not guarantee" something the report did not complete before directing readers to the full story.
The report does not detail how the watermark is detected, when the EU rollout began, or whether it will expand to other regions. No independent confirmation of the performance claims is included in the available material.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
An invisible, machine-readable watermark in text output is rolling out to ChatGPT and Codex, but only for users in the European Union at first. 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 […]