Sakhanda Wire
NVDA $153.78 -1.51% MSFT $504.71 -2.56% GOOGL $279.10 -1.63% META $617.85 -1.37% AMZN $256.64 -0.43%
← Back to the news

Anthropic explains how Claude’s text watermarking will work

Anthropic has provided more details about its plan to watermark text generated by Claude. The company says the system will encode a pattern through low-stakes word choices, such as selecting one of several similarly suitable descriptions. The pattern should be invisible to readers but detectable with the appropriate key, and Anthropic says it will not affect the quality or appearance of responses.

The company plans to use the SynthID-Text approach described by Google DeepMind in 2024 and release an API for detecting watermarks. Anthropic distinguishes this method from AI detectors that search for stylistic patterns associated with machine-generated writing. It says light editing is unlikely to remove a watermark completely, while replacing every word in a full rewrite will do so. In that situation, Anthropic said, it may be debatable whether the result remains AI-generated.

Watermark detection in text lightly proofread by Claude will depend on the passage’s length and the extent of the edits. Code is expected to carry less of a watermark because functional code offers fewer equivalent choices, although comments and other arbitrary wording may be marked. Anthropic said other major model developers that signed the EU AI Act’s Transparency Code are also planning watermarks.

Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.

Originally published by TechCrunch on

Read the original on TechCrunch ↗

The reporting and the image belong to TechCrunch and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news