Anthropic Introduces New Watermarking Method for AI-Generated Text to Enhance Transparency
Anthropic, the company behind the large language model Claude, is implementing a new watermarking technique for AI-generated text. This method involves biasing the AI's next word choice based on a secret key value, creating a pattern of favored word selections that can be detected by specialized software but remains imperceptible to human readers. According to Hong-Sheng Zhou, Ph.D., an associate professor in the Department of Computer Science at Virginia Commonwealth University, this approach is similar to cryptographic watermarking used for copyright protection in digital media. The goal is to introduce a detectable 'fingerprint' in AI-generated content without altering its meaning or quality. This initiative is partly driven by regulations like the European Union's Artificial Intelligence Act, which aims to increase transparency regarding AI's involvement in content creation. The watermarking relies on secret cryptographic information controlled by the model developer, meaning detection tools are also m...
