Summary:
- Anthropic has implemented a cryptographic watermarking system for text generated by its Claude AI models to improve the detection of machine-generated content.
- The mechanism works by subtly influencing the probability distribution of token selection during text generation, allowing the company to verify if a specific output originated from their systems.