Gaming News

AI-Generated Text Will Soon Be Detectable Thanks to an 'Invisible' Watermark

August 12, 2026Diego Herrera1 мин

Distinguishing between human-written and AI-generated text is becoming increasingly difficult. However, the European Union is implementing measures to address this, particularly concerning transparency and legal obligations. Anthropic, the creator of Claude, has announced that its new AI models will embed watermarks in AI-generated content right from their launch. Specifically, Claude models released in the EU from August 2, 2026, will support machine-readable watermarking. Generated texts will include embedded invisible watermarks, and supported files will contain digitally signed metadata for provenance. Anthropic is also working to integrate this watermarking into older Claude models.

Essentially, texts produced by Claude will feature an invisible watermark. Anthropic explains that this watermark is integrated directly into the text itself, without affecting its meaning, quality, or readability. It can even persist when the text is copied and pasted elsewhere. However, concerns about the quality of AI-generated text and potential methods to circumvent the system have already been raised, notably by Alex Cui, CTO of GPTZero.

While the system is designed to ensure transparency, it's not infallible. Anthropic emphasizes that detection will merely indicate that the content may have been processed by Claude, without definitively proving authorship. The watermarks can also become undetectable or disappear if the text is heavily edited, paraphrased, translated, combined with other texts, or if a passage is too short.