Anthropic Implements Watermarks in AI Texts

Anthropic and the New Era of Watermarks in AI Texts

Anthropic, known for its AI models like Claude, announced that it will implement watermarks in texts generated by its tools. This decision is not just a strategic choice but a direct response to the requirements of the new Transparency Code of the European Union, which came into effect on August 2. The idea is simple: any content generated or edited by AI must be marked in a way that other systems can easily identify it.

From now on, all models released by Anthropic after this date will come with this technology embedded. This applies to both texts and files, with the company adopting the open standard C2PA for the latter. And it doesn't stop there. Anthropic also plans to extend this functionality to older models, ensuring that the watermark accompanies the text even when it is copied and pasted elsewhere.

The Persistence of the Watermark

An interesting detail is that the watermark is an integral part of the text. This means that even if you copy and paste the content elsewhere, the watermark remains. And, in many cases, it can even survive some edits. Anthropic has made it clear that this marking will be applied at the model level, meaning it will be present in any product or platform that uses Claude.

But the question remains: how much can the text be edited before the watermark disappears? Anthropic has not fully clarified this point yet but promised updates as soon as it has more information.

The Race for Transparency

Anthropic is not alone in this race. Other companies, such as Google, Meta, Microsoft, and OpenAI, are also moving to comply with the European code. The pressure comes not only from regulations but also from users themselves, who demand more clarity about what is generated by AI and what is not. Recently, the music platform Suno announced that it would start marking its AI-generated tracks after facing legal challenges. And Substack, in partnership with Pangram, is also signaling AI-generated content.

The watermark is more than a legal requirement. It is a response to the growing demand for transparency and ethics in the use of AI.

The Impact for the End User

For the average user, this change may seem subtle, but it has profound implications. Knowing that a text or a song was generated by AI can influence how we consume and interpret that content. Furthermore, the watermark can serve as an additional layer of protection against plagiarism and misuse of AI-generated content.

In the end, the implementation of these watermarks is an important step towards a more responsible and transparent use of artificial intelligence. And as more companies adopt this practice, we can expect a clearer and more ethical market, where the origin of the content is as important as the content itself.

Anthropic and other tech giants are laying their cards on the table. And, if it depends on them, the future of AI will be marked, literally, by a new era of transparency.