The BMJ Group and the AWS Generative AI Innovation Center built an Editorial AI Assistant that helps editors screen submitted manuscripts, giving structured and sourced analysis while editors keep the decisions. It uses eight specialised agents covering areas such as scope, novelty, methodology, ethics and author integrity, running on Amazon Bedrock AgentCore with Claude models from Anthropic. BMJ Group’s nearly 70 journals receive tens of thousands of submissions a year, and about half are rejected at initial editor screening.
Benefits & outcomes
- The authors report over 80% success across three technical evaluation dimensions in a pilot on 21 manuscripts with editors from Emergency Medicine Journal and BMJ Open, the dimensions being flagging known errors, detecting fundamental methodological flaws and matching actual publishing outcomes.
- Editors in the pilot rated the methodology agents the most helpful, above 3.0 on a scale of 1 to 4, and the scope and ethics agents the least, at around 2.0.
Notes & quotes
- The authors say the agents did best against defined standards such as reporting checklists and were less reliable where experience-based judgement was needed.
- The authors list further testing and integration with the ScholarOne manuscript system as next steps.
- “AI can be a powerful ally in this domain, but only if we approach the opportunity with appropriate humility.” Ian Mulvany, Chief Technology Officer, BMJ Group.
Publisher / company: Amazon Web ServicesAnthropicBMJ Group
Date of case study: 2026
Status: Pilot
Built by: Vendor working with a publisher
Sources: