As autonomous AI systems grow more adept at research, scientists are grappling with a new competitive threat: being scooped by bots. Recent announcements from OpenAI and Anthropic—one on progress related to the Navier–Stokes problem and another on CRISPR-like DNA repeats in viruses—sparked concerns that proprietary models might benefit from users’ unpublished prompts, even as both companies deny training on such data. The dust-up is prompting labs to curtail use of commercial AI tools, citing intellectual-property risk and unclear data-handling practices. Some AI experts contend the core issue isn’t data leakage but that systems will increasingly outpace humans in problem-solving, raising pressure on publishers, funders and companies to clarify attribution, transparency and acceptable AI use in the scientific workflow. The resulting trust gap could reshape how discoveries are credited and how quickly researchers share work-in-progress.
Related articles:
Nature journals update policy on ChatGPT: tools cannot be listed as authors
Highly accurate protein structure prediction with AlphaFold
DeepMind’s AlphaDev discovers faster sorting algorithms
