DRYLAB-Bench: Evaluating biosecurity risks of agentic LLM mutation design
Proposing DRYLAB-Bench to evaluate agentic LLMs' high-risk mutation design capability
- arXiv
- 2610.13116
- Published
- Layer
- Application layer
- Setting
- science agent
- Evaluated models
- Claude Opus 5, Claude Opus 4.8, Claude Fable 5 and others15