Anthropic adds watermark to Claude, sparking debate over AI text authenticity

Anthropic rolled out a watermark feature in Claude that subtly modifies generated text so that machine detectors can flag it, while the changes remain invisible to human readers [Daring Fireball].

The addition targets a growing problem: AI‑generated prose is becoming indistinguishable from human writing, and existing detection tools are struggling to keep pace.

Critics describe the watermark as a “perversion of writing,” arguing that embedding machine‑readable markers blurs the line between author and algorithm and could erode trust in online content [Daring Fireball].

If adopted widely, the watermark may become a standard element of detection pipelines, forcing downstream tools to adjust or risk missing AI‑generated material.

Poll: Do you think AI‑generated text should be watermarked to promote transparency?

• Yes, it's necessary for authenticity
• No, it undermines creative freedom
• Only in specific contexts, like academic writing