Users are canceling their Claude subscriptions because of new AI flagging

AI company Anthropic has introduced invisible watermarks for its text generator Claude. In order to meet EU requirements, the system now marks generated content. Many developers fear disadvantages and cancel their subscriptions.

Trouble with invisible AI markers

A few days ago, the AI ​​company Anthropic introduced invisible watermarks for its text generator Claude. The aim of the measure is to comply with European transparency requirements for artificial intelligence. However, the technical implementation is met with massive criticism from many users, as the markings also remain in texts used for professional purposes or in the program code. The technology is based on the SynthID process. The system manipulates the probability of certain word sequences using a secret key. The result remains inconspicuous for human readers. However, appropriate software for recognizing AI texts can reliably identify the pattern. According to the developer, this works even if Claude corrects raw text written by humans.

Users fear professional disadvantages

Numerous software developers see the innovation as a significant risk to their work. Reports Business Insider as a result, a number of customers have ended their paid subscriptions to Claude. Users fear that secret markings in delivered code could lead to legal problems. A main point of criticism from users is the gradual loss of control. If AI software is used purely as an assistance tool for troubleshooting, it still leaves a digital fingerprint on the document. This poses a problem for service providers who are contractually obliged to provide their own work. Anthropic counters that a watermark only proves that Claude edited it, but not the complete authorship.

Google chooses a different path

While Anthropic makes marking mandatory for all users, other providers announce more flexibility. In the future, Google will allow its users to completely remove watermarks from generated texts and media. In doing so, the company is responding directly to the concerns of many developers and copywriters who fear disadvantages from compulsory marking. The current debate highlights the growing tension between regulatory requirements and the needs of end users.

Proponents of watermarking emphasize the need to reliably label AI content on the internet. Critics, on the other hand, demand adaptability, especially when they use the tools for editorial tasks. Whether the protests lead to a policy change at Anthropic remains to be seen.