The Invisible Mark - Who Really Created This?

What happens when artificial intelligence leaves an invisible mark on the things we create?

Anthropic has introduced a new approach to identifying content generated or processed by Claude, using invisible text watermarking and content provenance technologies. At first glance, this sounds like a welcome response to deepfakes, synthetic media, misinformation and the growing difficulty of knowing where digital content actually came from.

But there is a much more complicated question hiding underneath.

If you spend hours researching and writing an article, then ask Claude to correct the grammar, who created it? If a student develops an original argument but uses artificial intelligence to improve the writing, is the assignment artificial intelligence-generated? If a photographer captures an original image but uses artificial intelligence to enhance it, does the technology deserve part of the authorship?

In this episode, I explore Anthropic's move towards marking Claude-generated content and why digital provenance could become an important trust layer for the internet.

There are significant benefits. Journalists could gain another tool for examining suspicious media. Businesses could better understand where artificial intelligence has entered important documents and workflows. Creators may have stronger mechanisms for establishing provenance. Organisations could develop clearer audit trails, while entirely new entrepreneurial opportunities may emerge around authentication, verification, intellectual property and content provenance.

But there are risks.

A watermark can indicate that artificial intelligence touched something without proving that artificial intelligence created the underlying idea. That distinction could have serious consequences for students, writers, employees, job applicants, artists and entrepreneurs if detection becomes confused with authorship.

We could also see an emerging battle between provenance technologies and tools designed to remove their fingerprints.

Perhaps the future will not be divided neatly between "human-generated" and "artificial intelligence-generated" content at all.

Instead, we may need to recognise different levels of contribution: human conceived, artificial intelligence assisted, artificial intelligence transformed, human reviewed and human approved.

Because as artificial intelligence becomes embedded in everyday creation, the important question may no longer be whether a machine touched our work.

It may be something far more fundamental:

What did the human contribute?