Anthropic recently clarified its strategy for identifying AI-generated content, confirming that it will utilize a version of Google DeepMind’s 'SynthID-Text' technology. This system works by subtly adjusting the probability of word choices in Claude’s output, creating a pattern that remains invisible to human readers but can be detected by specialized software.

The implementation of these watermarks is a direct response to the European Union’s AI Act. This regulation mandates that companies must clearly label synthetic content, including text, images, and audio, to prevent misinformation and increase transparency. In addition to the text watermarking system, Anthropic is also incorporating C2PA standards for images processed by Claude. C2PA is an open technical standard that provides a verifiable history of digital media, allowing users to see if an image has been altered or created by AI.

While the technology is currently being rolled out to meet EU requirements, the broader implications for AI safety and content moderation remain a subject of industry discussion. By adopting an open-source framework like SynthID, Anthropic is aligning itself with a growing trend among major AI labs to standardize how synthetic content is tracked across the internet. The company has not yet provided a specific timeline for when these features will be fully integrated into all user-facing versions of Claude, but the announcement confirms that compliance with international transparency laws is a primary driver for the update.