Anthropic has integrated Google DeepMind’s SynthID-Text watermarking technology across all its Claude models. Starting August 2, 2026, every new Claude model will carry a statistical fingerprint invisible to the human eye but identifiable through a cryptographic key. This measure, applied globally and not limited to Europe, aims to comply with EU AI Act Article 50(2) Code of Practice. Existing Claude models will be retrofitted by December 2, 2026. Google’s deployment of SynthID-Text on Gemini, validated across roughly 20 million responses, confirms no measurable quality degradation. However, recent research raises questions about potential side effects on tool selection and adherence to safety guardrails.
Source: Read the original article
