Claude to Add Watermarks, and More AI Publishing News

3 hours ago 1

We’re turning some kind of corner in the ongoing conversation about if, where, and how to deal with the reality that AI-generated text has already made its way into mainstream publishing.

Here’s a look at the biggest recent headlines:

Welcome to the uncanny valley. In the most meta (but not Meta) of experiments, Mother Jones‘s David Corn used Claude to generate a novel about a writer who used Claude to generate a novel. It took a couple tries, but the ultimate result was “not so bad.”

Though flawed, The Ghost in the Machine (Claude’s got jokes) kept Corn engaged: “I wanted to keep reading. The plot is taut and, though I knew what was coming, the journey to the narrative guideposts I requested is a good ride, with a few surprises along the way.”

Thanks, EU. The same international body that forced Apple to ditch the lightning port has now compelled AI companies to “add machine-readable marks to enable the detection of AI-generated or manipulated content.”

Anthropic was the first tech company to respond publicly, announcing last week that it will add watermarks to text generated and edited by its large language models (LLMs), including Claude.

Who detects the detectors? New York Times tech journalist Brian X. Chen put popular AI-detection program Pangram through its paces, concluding after dozens of tests that “its proficiency in detecting bot-produced words will make it very useful for sniffing out annoyances” like scam emails, LinkedIn slop, and fake online reviews.

Substack has also bet big on Pangram’s powers with a new tool that allows creators and readers to scan text on the platform for AI-generated content.

An ounce of prevention is worth a pound of cure, which is why writer Vera Kurian has developed an app called to help authors prove their work is not AI-generated.