The Provenance Paradox: Why Anthropic's Watermarks Ignore the Creator Payday
AI-generated image · Bay Street Wire
Anthropic is deploying SynthID-Text to satisfy EU regulators, but marking the output does nothing to solve the systemic monetization crisis for the humans fueling the models.
Anthropic is moving forward with a plan to watermark text generated by its Claude chatbot, a decision the company says is necessary to comply with the Transparency Code of the EU AI Act, as TechCrunch first reported. The company intends to use the SynthID-Text approach—a method developed by the Google DeepMind team in 2024—to embed patterns in AI responses that remain invisible to readers but detectable via a specific key.
From a technical standpoint, Anthropic describes this as a series of "low-stakes choices," where the model selects between synonymous words (such as "grey" versus "overcast") to create a detectable signature. The company asserts that this process does not degrade the quality of the output and plans to release a watermark detection API to facilitate identification.
However, this focus on provenance is a distraction from the central crisis facing the creator economy: the monetization of data. While Anthropic focuses on whether a piece of text is AI-generated, it offers no solution for how the human creators whose work fuels these models are compensated. Watermarking is a tool for identification, not a mechanism for remuneration. It tells us where the text came from, but it doesn't ensure the original architects of the knowledge are paid.
Even the efficacy of this "solution" is questionable. TechCrunch notes that Anthropic admits a complete rewrite of the text would remove the watermark entirely. Furthermore, the watermark's presence in code is described as "negligible" because the model lacks the freedom to make arbitrary word choices in functional programming, though it may appear in code comments.
***
**Opinion:** In my view, the industry's pivot toward watermarking is a desperate attempt to appease regulators without addressing the underlying economic imbalance. By focusing on the "transparency" of the output, platforms avoid the harder conversation about the value of the input. If the goal is truly to protect the information ecosystem, the focus should shift from marking the AI's work to compensating the human labor that makes that AI possible.
***
Despite the technical justifications, the move has already sparked a backlash. TechCrunch reports that users on Reddit have characterized the move as a "conspiracy," while Business Insider reports that "dozens" of users on X have claimed they are canceling their Claude subscriptions. This friction highlights a growing divide: platforms are prioritizing regulatory compliance and corporate safety over the autonomy and economic security of the users and creators they rely upon.