Anthropic has launched a new tool designed to help detect text generated by artificial intelligence by embedding an imperceptible watermark in content produced by its Claude models.
The company said the watermark does not affect the meaning or readability of the text and remains embedded when the content is copied and pasted. The feature will be applied to Claude-generated content worldwide, including content produced through cloud service providers, with Anthropic planning to provide third-party organisations with tools to detect the watermark.
Claude models launched this month and subsequently will support the feature from the outset, while Anthropic is working to introduce it to older models. The initiative forms part of the company’s commitment to greater transparency under the European Union’s Artificial Intelligence Act.
The watermark will provide publishers, educational institutions and other organisations with an additional means of verifying whether AI was used in written content, at a time when the publishing industry is facing growing challenges associated with AI-generated material.
Anthropic cautioned, however, that extensive editing, rewriting, translation or combining Claude-generated content with other writing may render the watermark undetectable. The presence of the watermark also does not conclusively establish that Claude originally generated the text.
New System Enables AI to Conduct Scientific Experiments in Laboratories
In a separate development, Anthropic has unveiled a research version of a system that enables AI agents to control scientific and engineering equipment and conduct experiments, extending the use of its technology beyond the digital environment and into physical laboratories.
Known as the Model Context Protocol for devices, the system enables AI agents to operate and coordinate a wide range of programmable equipment, including microscopes, liquid-handling systems, lasers and robotic arms. It can be used to conduct experiments across fields such as drug discovery, robotics and quantum computing.
Anthropic is testing the technology with a number of laboratories and equipment manufacturers. Initial trials have demonstrated that AI agents can monitor experiments, modify procedures and repeat experiments, performing tasks that resemble some of the activities carried out by scientists.
The company plans to make the system open source once appropriate safeguards have been established to mitigate potential errors and risks, particularly in applications related to the life sciences.
The launch follows Anthropic’s introduction of Claude for Science, a product designed for researchers, as the company continues to expand the use of artificial intelligence in scientific research, protein design and drug discovery.
