Anthropic has begun embedding invisible, machine-readable watermarks into text generated by its Claude artificial intelligence models, introducing a hidden signal that can remain with AI-generated content after it is copied and may survive some forms of editing.
The company disclosed the development on Monday, saying the marking system forms part of its commitments under the European Union’s AI Act Code of Practice on Transparency of AI-Generated Content.
Anthropic said Claude models launched on or after August 2, 2026, support machine-readable marking from launch, while the company is working to extend the capability to models released before that date.
The company said the technology is intended to provide greater transparency as AI-generated content becomes increasingly widespread and give users additional context about the origin of information they encounter.
For text, Claude embeds the watermark directly into its generated output rather than attaching a separate label or file.
Anthropic said the watermark is designed to be imperceptible and does not alter the meaning, quality or readability of the generated response.
The company also said the marking can travel beyond Claude when text is copied and pasted elsewhere. Depending on the extent of subsequent changes, the signal may remain detectable after some editing.
Anthropic is developing systems that will allow users and third parties to detect Claude’s markings. The company said more details about how the detection mechanisms work will be provided in forthcoming technical documentation.
However, Anthropic cautioned that detecting a Claude watermark should not be treated as definitive proof that Claude originally authored a piece of content. Users may employ the AI system to perform tasks such as proofreading, translation, summarisation or file conversion without having Claude generate the original work.
Similarly, the absence of a detectable watermark does not prove that content was produced entirely by a human.
Anthropic said the markings can become undetectable when text is heavily edited, paraphrased, translated or combined with other writing. Very short passages may also lack enough material for reliable detection.
The development comes as the rapid adoption of generative AI continues to raise questions about the amount of work being produced or significantly assisted by artificial intelligence, particularly in education.
In Nigeria, university lecturers have raised concerns about students increasingly relying on AI tools such as ChatGPT to complete assignments and answer tests.
Prof. Isaac Nwaogwugwu of the University of Lagos told Nairametrics that about 40 percent of more than 100 MBA students in one of his classes submitted identical answers after apparently using the same AI tool.
Concerns have also extended to lecturers. Dr. Ereke of Ebonyi State University told Nairametrics that some lecturers were using AI to generate lesson notes, course outlines, marking schemes and examination questions without adequately reviewing the material.
The emergence of machine-readable AI markings could therefore become particularly relevant to educational institutions. If students use generative AI to produce assignments that appear to be entirely their own work, watermark detection could eventually provide schools with an additional signal when assessing whether submitted material has been processed by an AI system.
The technology does not, however, provide a definitive solution to AI-related academic misconduct. Watermarks may disappear following extensive editing or other transformations, while legitimate uses of AI, including proofreading and translation, can make it difficult to determine precisely how much of a piece of work was generated by an AI system.
Nigeria is also moving to incorporate artificial intelligence into its education system, with the government positioning the technology as a tool for improving teaching and learning while maintaining human agency.
The Minister of Education, Tunji Alausa, previously said the ministry had integrated AI-powered platforms providing information on tertiary institutions and admissions across Nigeria as part of efforts to incorporate artificial intelligence into teaching and learning.
As AI-generated content becomes more common across schools, workplaces and online platforms, technologies such as Anthropic’s watermarking system could play a growing role in helping users understand the provenance of digital content while highlighting the limits of determining who ultimately created it.
