Anthropic says its Claude AI cut one drugmaker's document review from over 25 hours to under three, and it is now working on other steps of drug development.
Anthropic says its AI system, Claude, helped a biotech company called Enveda cut the time to check a big stack of manufacturing paperwork by 89%. The review went from more than 25 hours to under three. Anthropic described the results in a new blog post about its work with drug developers.
Enveda used Claude on manufacturing documentation. That paperwork had taken up to 75% of some chemists' working time. In one test, the system read 1,061 pages. It found all 11 problems that human reviewers had already caught. It also found seven more.
The paperwork is only one part. Anthropic says Claude is being used at several stages, from designing molecules to planning clinical trials. The projects cover cancer, food allergies and rare muscular disorders.
Most of these treatments are still experimental. None are available to patients yet. The results also come from Anthropic's own account and the reported findings of its partners. They are not an independent review.
Michael Pourdehnad of Bristol Myers Squibb said Claude produced a robust first analysis from a single prompt. He said it helped weigh options that would otherwise have taken several days.
The Enveda result is the easiest to relate to. Checking long documents for errors is slow work in almost every company. The example suggests AI can take on a large share of that first pass. Note that in this case humans had already found the original 11 problems, which gave a way to check the AI's work.
Anthropic and Bristol Myers Squibb plan to explore how Claude could help with decisions in early clinical development. Anthropic also plans more partnerships, research programs and tools for life-science organizations.
AI may speed up routine document checks, but keep a person in the loop to verify what it finds, as the Enveda test did.
World models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasksA University of Chicago law professor asked Claude to fix page numbers. It deleted his syllabus and a year of notes instead. Dropbox saved him, not the AI.
Read more: An AI assistant deleted a law professor's year of lecture notes
