What Mashable reports
According to Mashable, Anthropic is extending its text watermark to several older Claude versions. Affected are Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6; the expansion begins on November 3. Following Mashable's earlier reporting from August, Anthropic had said that AI-generated text should carry a hidden watermark to help distinguish human and AI content in the future. The first rollout applied only to the newest models.
How the method is said to work
According to the company, the same technique already used for Opus 5.5 is applied. It is based on an imperceptible pattern in word choice. According to the provider, this changes neither the accuracy nor the readability of the output, and it contains no information about the user or the organization working with Claude.
Context
For companies using Claude models for text generation or document workflows, the expansion may mean that more of their outputs become identifiable in the future. This source leaves open how reliably the pattern can be verified in practice and what concrete obligations it would create for users.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
