Anthropic is preparing an API to detect Claude's texts

⚡

Key Takeaways

1The detection approach has limitations with factual texts, code, and rewrites.

2Anthropic plans to release an API for third parties to verify if a text comes from Claude.

3The method relies on Google’s SynthID and adjusts randomness when selecting words.

💡Why it matters — The announced API is expected to provide third parties with a way to verify the Claude origin of a text.

⚡Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄

Full Analysis

The approach has limitations for highly factual texts, code, and substantial rewrites. Anthropic plans to provide an API that allows third parties to verify if a text originates from Claude, relying on Google’s SynthID and adjusting the randomness during word selection, a method the company claims does not degrade quality.

Announced limitations for facts, code, and rewrites

The approach is noted to be less reliable for fact-rich texts, for code, and in cases of significant rewrites. These reservations limit the scope of detection.

The chosen method relies on SynthID and modifies word selection

The technology being considered is based on SynthID, a method developed by Google. It adjusts the randomness during word selection. Anthropic states that this modification does not harm the quality of the text.

An API for third parties announced and expected soon

Anthropic has announced a watermark detection API that will be offered soon. It is intended to allow third parties to verify whether a text was generated by Claude.

⚡

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.

📰 Voir toutes les actus IA →