EU AI Act warning says AI text may need machine-detectable marks
From August 2, 2026, the EU AI Act is claimed to require AI-made text to be marked in a way machines can read and detect. The main point is that a simple label saying something was made by AI may not be enough. AI-made audio may need metadata tags and fingerprint-style watermarks, with large fines possible for violations.
Providers of AI systems that generate text may need to make outputs and detectable as , using methods that are effective, interoperable, robust, and reliable. Very simple may be exempt, but models classed as systemic risk general-purpose AI may not be. Qwen 3.6 series, , GLM, and Kimi are named as examples in that risk category.
If ChatGPT or a similar tool offers a PDF or TXT download, the file may need , and the claim is that two layers of detection are required rather than one.
Key points
- The claim centers on an August 2, 2026 EU AI Act requirement.
- text may need to be and detectable as AI-made.
- A plain “AI generated” label may not be enough on its own.
- PDF or TXT exports may need .
- may still be affected if they fall under systemic risk rules.