Anthropic, an AI model maker, has announced that it will implement watermarking for the text generated by its models, adhering to new European regulations. The update on its support page highlights that all new models released after August 2 will feature technology capable of embedding watermarks in computer-generated text and files. This move follows the implementation of the EU AI Act’s Transparency Code, which mandates that AI-generated or edited content be distinctly marked for identification by other systems.
The watermark technology applied by Anthropic will also adhere to the C2PA open standard for files, ensuring consistency and recognition of the watermarked content across different platforms and systems. This watermarking will be automatically included in various products from Anthropic, including the Claude platform API, Claude, Claude Code, Claude Cowork, and Claude Tag. Furthermore, the watermarks are designed to persist when the content is copied and pasted, and may remain through certain levels of editing, although the degree to which editing might remove the watermark was not specified in the announcement.
The company’s decision to include watermarking extends to older models as well, indicating Anthropic’s commitment to comply with the new regulatory standards across its entire range of products. The statement on their support page explains that watermarking will be applied at the model level, ensuring that any text generated will carry the watermark regardless of the specific product or interface used.
This proactive approach by Anthropic is in response to increasing pressure on platforms that generate AI content to ensure transparency and accountability in their outputs. The need for watermarking has been accelerated by user backlash and regulatory scrutiny, which have pushed other companies in the sector to adopt similar measures. For instance, the AI music platform Suno recently announced it would mark tracks created on its platform following legal challenges, and the newsletter service Substack, in collaboration with Pangram, introduced flags for AI-generated content.
Watermarking by AI companies is seen as a crucial step in addressing the challenges posed by the pervasive use of AI in content creation, which can sometimes include deceptive practices like “Claudefishing”—a term mentioned by the CEO of Substack for deceptive content generation using AI. Transparency in AI-generated content is crucial not only for complying with legal standards but also for maintaining user trust and integrity within digital ecosystems.
The broader industry movement includes major tech companies such as Google, Meta, Microsoft, OpenAI, and Synthesia, which have all committed to adhering to the EU’s Transparency Code. This collective momentum highlights the growing recognition of the importance of ethical standards in the deployment of AI technologies, particularly those capable of creating or modifying content that could potentially mislead consumers or influence public opinion.
The implementation of such technology also reflects a growing understanding of the need for AI-generated content to be identifiable, ensuring that users and regulators can distinguish between human and machine-created texts. This is increasingly important in a world where AI-generated content is becoming more sophisticated and widespread, raising concerns about the potential for misinformation, copyright issues, and other legal issues.
As companies like Anthropic take steps to incorporate these regulatory requirements into their operational protocols, the AI industry is moving towards greater transparency and accountability. However, the effectiveness of these measures in curbing abuses and maintaining public trust in AI-generated content will likely continue to be a subject of discussion and refinement in the regulatory and technological landscape.
Read the full post on techcrunch.com
