▲ Anthropic
When Anthropic announced plans to introduce watermarks for text AI outputs, some users strongly protested, with some even threatening to cancel their subscriptions.
In response, the company stepped in to calm users, explaining that the implementation of watermarks is a mandatory requirement under European Union law and applies to other companies as well.
Social media platforms like X and Reddit were flooded with concerns regarding Anthropic's introduction of identification markers.
Users expressed outrage, calling the measure "a conspiracy targeting genuine Claude users," and stated their intention to cancel their paid Claude subscriptions.
Some worried that even if they used Claude solely to proofread and edit text they wrote themselves, the entire piece could be classified as AI-generated and result in disadvantages.
Experts also pointed out that complex issues such as copyright could arise.
On the other hand, some users countered the backlash, arguing that those objecting to the measure "were doing so because they wanted to deceive others."
As the protests continued, Anthropic explained via its website that identification markers will eventually be introduced not only for Claude but also for AI models from other companies.
Anthropic elaborated, stating, "Starting August 2, the EU mandated that AI companies providing services in its market must indicate that content is AI-generated," and added, "Other major developers also plan to introduce their own identification markers."
Along with this, Anthropic introduced the specific method it adopted in detail to alleviate user concerns.
The company explained that because it determines whether content is AI-generated by comparing the word-choice pattern across the entire text against Claude's patterns, text that users wrote themselves and merely had modified by AI will not be judged as AI content if only minor edits were made.
The company also explained that even for text generated by AI, if users substantially overhaul it, the AI identification pattern is unlikely to be recognized.
Anthropic added that it will also release a tool allowing users to check whether content is AI-generated themselves to ensure transparency.
(Photo: AP, Yonhap News)
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.