Anthropic's Claude cut Enveda's drug records' review time by 89%

Artificial intelligence (AI) company Anthropic said its model Claude helped Colorado-based biotechnology company Enveda reduce the review time for drug manufacturing records by 89%.

The company said in its blog post Enveda built a system to connect its manufacturing documents to Claude, allowing it to flag potential issues in records for demonstrating that drug batches meet quality standards.

While Enveda’s reviewers took 25.9 hours to manually examine the records, the AI system – CMC Document Processor (CDP) – along with human verification took 2.8 hours, cutting the review time by 89%, as per the report.


“These projects offer promising first signs that AI tools might speed up the process of document review, which is one of the bottlenecks to getting a new medicine approved,” Anthropic said.

How Claude helped Enveda

A manufacturing chemist at Enveda could spend roughly 75% of their time reviewing documentation, Anthropic said. In one instance, a senior reviewer needed 10 hours to assess a single 500-page record from a contract manufacturer.

Enveda developed the CDP to help reviewers identify errors, inconsistencies and potential quality issues across these records. The system compares manufacturing documents against relevant instructions and specifications, flagging issues for human reviewers to assess.

During the evaluation, the system identified each of the 11 issues that human reviewers had flagged. In the timed run, it also identified seven additional issues that Enveda's quality team said were required to be addressed before the batch was released.

These included a manufacturing step that exceeded its permitted time limit and a test marked as failed despite having no recorded result.

“Enveda reports that after three internal hackathons, most of its technical team can now generate first drafts of [Investigational New Drug] content themselves, which the company says gives them more time to prepare other regulatory submissions and develop protocols for additional programmes,” Anthropic added.

However, the AI system did not replace human oversight. Anthropic said Enveda's quality team remained responsible for reviewing the findings and deciding whether to release a batch.

Cutting short the time

The collaboration also explored using Claude to prepare regulatory documents required before a drug can be tested in humans.

As per Anthropic, Claude generated the first draft of a section of an application submitted to the US Food and Drug Administration (FDA) in four hours, compared with approximately three months needed to do it manually. The estimate excludes the three-week review process required for both approaches.

Enveda develops medicines using molecules produced by living organisms. Since 2019, the company has produced 17 drug candidates, three of which are in clinical trials, according to Anthropic.

Both Anthropic and Enveda said the findings suggest AI could help reduce the administrative burden of drug development but the results are from specific workflows and do not establish that AI can independently manage pharmaceutical quality checks or regulatory submissions.