Research Operations, External Artifacts

About the role

Anthropic publishes risk reports: long-form technical documents laying out our assessment of the most serious potential risks from our models in domains like CBRN, cyber operations, and AI autonomy, along with the evaluation results behind that assessment, the safeguards we've applied, and our reasoning for why a given model is safe to deploy under our Responsible Scaling Policy. Some risk reports are standalone periodic assessments; others are more targeted, produced when we release a specific frontier model. These are some of the most consequential documents we produce, and one of the main ways we hold ourselves publicly accountable for the safety claims we make.

We're hiring a Research Operations Specialist to own risk report operations. You'll be embedded with safety and research teams through each report cycle: coordinating contributions from dozens of researchers, holding the schedule and the open-threads list, and making sure the document ships on time as a single, internally consistent whole. You'll also do substantive editorial work, turning evaluation results, threat models, and researcher notes into clear prose and pushing back when a safety argument doesn't hold together.

Risk reports sit within a wider family of external safety artifacts, including system cards and Responsible Scaling Policy updates. Part of this role is keeping those documents consistent with each other so that what we commit to in one place matches what we commit to and deliver on everywhere else.

This role sits in Research Operations and works closely with our Frontier Red Team, Safeguards, Alignment, and capabilities researchers. The job is part project management, part translation: keeping a complex, many-author, hard-deadline document on track while making frontier risk assessment legible to researchers, policymakers, journalists, and the public without losing precision.

Key responsibilities

• Drive risk report production end to end: own the timeline, the contributor list, and the open-threads tracker
• Coordinate core contributors across Frontier Red Team, Safeguards, Alignment, Interpretability, and capabilities research; chase drafts, resolve disagreements, find ground truth, and run the final polish pass
• Edit (and sometimes write) content; work with researchers and red-teamers to turn evaluation results, threat models, and plots into clear, non-marketing prose, and keep Anthropic's voice consistent across sections drafted by many different people
• Guard accuracy and consistency: catch terminology drift, risk claims that subtly contradict each other, and gaps between internal findings and what the draft says
• Keep the risk report aligned with system cards, RSP disclosures, and other safety documentation, and flag conflicts early
• Improve the process between reports; build templates, style guidance, and contributor checklists so each cycle starts from a stronger baseline
• Pick up other research-adjacent operations and writing work related to our external artifacts and Anthropic's RSP

Minimum qualifications

• Demonstrated technical writing ability: can take dense, jargon-heavy source material and produce prose that is precise and readable by a smart non-specialist
• Working conceptual knowledge of large language models, with fluency in terms like pretraining, RLHF, context windows, evals, red-teaming, and capability thresholds
• Ability to r…