AI glossary · Basic · AI security · Generative media
AI watermark
AI watermarking is a technology that embeds hard-to-notice marks in AI-generated text, images, audio and video so that it can later be checked whether something was 'made by AI.'
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Security & safety
- Fields
- AI security, Generative media
- Added
- 2026-10-08
How it works
For images and audio, subtle signals that are hard for human eyes or ears to notice are mixed into the pixels or sound waveform. For text, when the AI chooses the next word (token), it is made to pick certain words slightly more often according to a specific rule, leaving a statistical trace. Later, a detection tool checks for these signals or traces and reports the 'likelihood of AI generation.' In many cases, only the party that knows the rule (key) used to embed the signal can detect it accurately.
Why it matters
As AI-generated content becomes harder to tell apart from human-made content, problems such as fake news, deepfakes and impersonation have grown. Watermarks provide clues for verifying where content came from, which helps reduce this kind of harm. As regulations such as the EU AI Act start requiring AI-generated content to be labeled, companies have more reason to adopt watermarks.
Good to know
Watermarks are not a cure-all. Paraphrasing or translating text can weaken the trace, and cropping, compressing or re-photographing an image can damage the signal. The absence of a watermark also does not prove that a person made the content, because some AI services do not add watermarks. Detection results are usually given as probabilities, so the possibility of a wrong call should be kept in mind.
Example
Among the news this site has covered, there was a report that Google released a website worldwide that detects AI-generated content based on its own watermarking technology, SynthID, and that it also supports detecting content generated by other companies' tools. There was also news that OpenAI is applying watermarks to text generated by ChatGPT and Codex in response to EU rules. These cases show watermarks quickly making their way into real services alongside regulation.