Product Manager, Safeguards Rare Harms

Anthropic is dedicated to developing AI assistants that are helpful, harmless, and honest. The Safeguards team builds protections for new AI features and products. As a Product Manager, you will own the ideation, design, development, and deployment of Safeguards systems and relevant product UX to advance frontier models safely across various cloud platforms.

• Determine how to build in safety by design upstream and leverage downstream defenses for Anthropic’s frontier models, AI products, and customers on different surfaces (Claude.ai, 1P API, external Cloud providers).
• Write safety evals and communicate externally about safety.
• Drive impact via ruthless prioritization, defining problems, solution options, and clear requirements toward MVP vs. ideal state.
• Align and collaborate with policy, enforcement, research, engineering, and cross-functional stakeholders.
• Understand the AI landscape to plan for mitigation of deployment risks.
• Lead development of metrics to understand area, performance, and blindspots.

Minimum: Ability to make technical tradeoff decisions, ideally with experience across policy experts, AI/ML research engineers, and software engineering teams. Strong user understanding of products and Safeguards concerns. Demonstrated ability to build product and engineering strategy across cross-functional teams. Experience in designing and building metrics. Ability to navigate ambiguity and prioritize. Planning, building, launching, and measuring new products/systems in a zero-to-one environment. Clear communication of complex technical concepts.

Preferred: 5+ years in product management with focus on fast problem understanding, building roadmaps, and getting into details on data, detection & interventions, infrastructure & tools, and/or evals.

Annual Salary: $305,000—$385,000 USD.

Minimum education: Bachelor’s degree or equivalent. Location-based hybrid policy: expect all staff to be in one of our offices at least 25% of the time. Visa sponsorship is available.