Technical Program Manager, Safeguards - Launches

Technical Program Manager, Safeguards - Launches About Anthropic

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.

As a Technical Program Manager for Safeguards focused on model launches, you’ll be the critical bridge between Research, Inference, Product, and Safety teams during high‑stakes model and product releases. You’ll drive end‑to‑end execution of Safeguards deployment for model and product launches, ensuring our safety classifiers are properly configured, deployed, tested, and scaled across all platforms while balancing capacity constraints, performance requirements, and user experience considerations.

• Own end‑to‑end safeguards readiness for model launches – drive all safeguards deployment activities from pre‑launch planning through post‑launch stabilization, ensuring classifier systems are properly configured, tested, and scaled for each new model release
• Coordinate complex cross‑platform deployments – manage classifier rollouts across platforms, each with distinct infrastructure constraints, capacity requirements, and deployment timelines
• Make real‑time technical trade‑off decisions – balance competing constraints around classifier capacity, latency impacts, false‑positive rates, and robustness requirements during fast‑moving launch cycles with aggressive timelines
• Drive capacity planning and performance optimization – partner with Inference teams to model compute requirements, optimize classifier configurations (batch sizes, mesh topologies, chip allocation), and ensure we stay within overhead budgets while meeting safety requirements
• Lead launch‑day war‑room coordination – serve as primary safeguards point of contact during launches, coordinating monitoring, troubleshooting classifier issues, and driving go/no‑go calls on deployment decisions
• Build detailed execution plans and run‑of‑show – create comprehensive launch checklists covering classifier configurations, monitoring thresholds, alerting setup, rollback procedures, and cross‑team dependencies
• Partner with product teams on co‑launch coordination – ensure new product surfaces have appropriate safeguards protections configured, tested, and deployed; whether launching independently or alongside new models
• Monitor and respond to false‑positive patterns – track flag rates across customer segments, identify problematic patterns, coordinate with research teams on classifier retraining priorities, and manage exemption processes for enterprise customers
• Manage transition to business‑as‑usual operations – once launches stabilize, coordinate handoff of ongoing monitoring and optimization work to sustaining engineering teams, then move to the next model or product launch as needed
• Maintain technical documentation – document classifier deployment configurations, capacity estimates, performance benchmarks, and lessons learned to build institutional knowledge

• Have deep technical program management experience in ML/AI systems – several years coordinating complex deployments involving model inference, distributed systems, and real‑time production services at scale
• Have experience with AI safety or Trust & Safety – familiar with concepts such as false‑positive/false‑negative trade‑offs, adversarial robustness, content moderation systems, or similar safety‑critical technical systems
• Can navigate high‑ambiguity technical challenges – synthesize incomplete information from multiple engineering teams, identify critical‑path blockers, and drive decisions when faced with competing technical constraints and tight deadlines
• Excel at rapid decision‑making under pressure – comfortable making high‑stakes trade‑off decisions with incomplete information during time‑critical launch windows, balancing safety requirements against capacity, latency, and user experience constraints
• Are skilled at cross‑functional coordination in complex technical…