Question to the Department for Science, Innovation & Technology:

To ask His Majesty's Government whether they were aware that Anthropic downloaded over 7 million pirated books between 2021 and 2023 to train its Claude AI model when selecting Anthropic as the supplier for the GOV.UK chat service; and whether they conducted a risk assessment for the use of a model developed using unlawfully accessed copyright-protected works.

The Government Digital Service recognises the importance of ethical, legal, and data protection considerations in the use of large language models.

A range of large language models from multiple suppliers have been considered in developing GOV.UK Chat. GOV.UK Chat is a product in active development that currently accesses Anthropic models through an existing agreement with Amazon Web Services, enabling the Government Digital Service to test a range of models from different suppliers. Risk assessments have been undertaken in accordance with government standards, including consideration of ethical, legal, and data protection risks relevant to the models tested.

GOV.UK Chat is being developed entirely within the Government Digital Service by a multidisciplinary team of civil servants. The Government continues to engage with UK-based AI developers and remains open to collaboration where this supports innovation and the delivery of public services.