Urgent.News

What's breaking now, across hundreds of outlets.

AI

OpenAI will watermark ChatGPT outputs by default—but only in the EU

Like other solutions, it is not especially reliable, and it's easy to circumvent.

OpenAI will watermark ChatGPT outputs by default—but only in the EU

OpenAI plans to automatically watermark ChatGPT outputs in the European Union, starting soon. However, the watermarking feature will be optional outside the EU for now. This initiative aims to comply with the EU AI Act, which went into effect last August. The law mandates marking AI-generated content in a manner that can be detected by other tools.

Despite ongoing efforts, there is no foolproof method to achieve this. Some standards, such as SynthID and the C2PA project, are easily bypassed by individuals with basic skills. OpenAI's own approach, named textGrain, remains a closely guarded secret. The company explains that it involves embedding patterns into word choices that are invisible to human readers.

While these changes do not significantly affect the output's quality, they can be identified using a specialized detector with the right key. OpenAI intends to share the detector with a select group of researchers and organizations, followed by a process for others to gain access in due time.

Rewritten by urgent.news; may contain errors — check the original.

Also reported by 1 other outlet

Every outlet that reported this, and when →

Read the original at arstechnica.com →

Report a problem with this story

More in AI

Getting to Reliable AI-Driven Development

To truly transform software delivery with AI, organizations must embrace a spec-driven approach. By grounding AI in clear specifications, this approach prevents AI from generating hallucinations while…

More from Tuesday 6 October →