Research
Anthropic research papers, surveys, and policy notes. This collection now includes 11 research articles, with new additions on coding agents, cybersecurity, AI leadership, open-source evaluation tools, and scientific benchmarking.
New in This Update
- Coding agents in the social sciences - Survey evidence on AI and coding agent adoption among quantitative social scientists.
- Project Glasswing: An initial update - Early findings from using Claude Mythos Preview to secure critical software.
- 2028: Two scenarios for global AI leadership - Policy analysis of AI leadership, compute, and global competition.
- Donating our open-source alignment tool - Petri 3.0 and the move to Meridian Labs.
- Focus areas for The Anthropic Institute - Anthropic's agenda for studying AI's real-world impacts.
- Evaluating Claude with BioMysteryBench - Benchmarking scientific and bioinformatics research capabilities.
Existing Research
- Teaching Claude Why - How AI models can understand reasons and values behind behaviors.
- Natural Language Autoencoders - Novel unsupervised learning via natural language representations.
- How People Seek Personal Guidance from Claude - User study on deep personal AI conversations.
- The AI Economy Told by 81,000 People - Large-scale survey on AI usage patterns and economics.
- Automated Alignment Researchers - Automating AI alignment research with AI.