Over the past month, two stories have emerged about AI companies publishing discoveries in the sciences. In both cases, academic researchers have claimed that these discoveries closely resemble their own unpublished work and that they had been actively using AI models as part of their research. This has led to suggestions that the AI models may have ingested data provided by researchers and that this information contributed to the reported breakthroughs.
The second story concerns Anthropic’s announcement of the discovery of new enzymes.
There are striking parallels between both stories and they raise important questions about the use of AI in research. They also highlight the importance of considering which AI services you use, what happens to the information you share with them, and what level of data protection is provided under the terms and conditions you accept when signing up to those services.
Durham University provides all staff with access to Microsoft 365 Copilot and ask.durham.ac.uk, both of which offer enterprise-grade data protection. Alternatively, staff can download LM Studio through Durham Apps Anywhere and run open-weight AI models locally on their own devices. These options can help researchers explore the benefits of AI while maintaining greater control over their data.