Staff+ Software Engineer – Safeguards Review Tooling

About the role

Anthropic is seeking a Staff+ Software Engineer to lead the development of review tooling for its Safeguards team. In this foundational position you will design and ship the internal platforms that enable safety investigators—and increasingly the Claude model—to examine potential harms, enforce policies, and feed insights back into model training across Anthropic’s products and third‑party cloud services.

Key responsibilities

• Design and build investigation, review, and enforcement tools, including case queues, investigation views, decision logging, and account‑action workflows for both first‑party and third‑party platforms.
• Create a reusable platform layer with APIs, data storage, and backend services that supports rapid, safe rollout of new review workflows.
• Drive automation of review processes, enabling Claude‑assisted and Claude‑driven workflows while keeping human judgment in the loop.
• Collaborate with policy, operations, legal, privacy, and data‑science teams to translate enforcement needs into reliable systems that reduce handling time and decision error.
• Implement robust guardrails such as granular permissions, audit trails, data‑access controls, and reviewer wellbeing features like content obfuscation.
• Instrument shipped tools to surface metrics on queue health, reviewer throughput, and overall system performance.

Required profile

• Proven experience building internal, safety‑critical platforms that support complex workflows.
• Strong ownership mindset, able to define product direction and deliver end‑to‑end solutions.
• Excellent collaboration skills with cross‑functional partners in policy, legal, privacy, and data science.
• Ability to balance rapid iteration with rigorous security and privacy requirements.

Required skills

What we offer

• Opportunity to shape core safety infrastructure at a leading AI research company.
• Work alongside top researchers, engineers, and policy experts.
• Competitive compensation and benefits package.