Research Operations, External Artifacts
Remote • Remote-Friendly, United States
Posted 16d ago
About the job
Anthropic is seeking a Research Operations Specialist to manage the production of risk reports, which are long-form technical documents detailing potential risks from AI models. This role involves coordinating contributions from numerous researchers, managing timelines, and ensuring the final document is cohesive and accurate. You will also perform substantive editorial work, transforming technical findings and threat models into clear prose, and ensuring consistency across various safety artifacts like system cards and Responsible Scaling Policy updates. This position requires a blend of project management and communication skills to make complex AI safety assessments accessible to a broad audience while maintaining precision.
Responsibilities
- Manage the end-to-end production of risk reports, including timelines, contributor lists, and tracking progress.
- Coordinate with researchers from various teams (Frontier Red Team, Safeguards, Alignment, etc.) to gather contributions, resolve disagreements, and finalize content.
- Edit and potentially write content, translating technical evaluation results, threat models, and plots into clear, consistent prose aligned with Anthropic's voice.
- Ensure accuracy and consistency in terminology and risk claims across the report and other safety documentation.
- Maintain alignment between risk reports, system cards, and other safety disclosures, flagging any conflicts.
- Improve operational processes by developing templates, style guides, and checklists for future report cycles.
- Undertake other research-adjacent operations and writing tasks related to external artifacts and Anthropic's Responsible Scaling Policy.
Requirements
- Demonstrated ability in technical writing, capable of producing precise and readable prose from dense, jargon-heavy material for a non-specialist audience.
- Working conceptual knowledge of large language models, including terms like pretraining, RLHF, context windows, evals, red-teaming, and capability thresholds.
- Ability to interpret evaluation results tables, ask clarifying questions, and identify weaknesses in technical arguments.
- Proven track record of successfully completing complex, multi-contributor projects with hard deadlines.