Engineering Manager, Safeguards Interventions

Anthropic•San Francisco, CA

13d•$405,000 - $485,000•Hybrid

About The Position

The Safeguards team is responsible for ensuring our models and products are developed and deployed safely. We're looking for an Engineering Manager to lead the Interventions team: the group responsible for what happens when a safety system fires. It owns the composable arsenal of systems that sit between our detection stack (classifiers and probes) and the user, across every Anthropic surface: 1P products, the API, and third-party clouds. This includes inline interventions for areas like bio, cyber, and acceptable usage as well as downstream areas like child safety and copyright. This team is responsible for ensuring that we evolve and drive the quality of our interventions to enable our products to grow safely.

Requirements

Have managed engineering teams shipping production ML or safety-enforcement systems where the system's decisions directly affected users.
Have run high-stakes, compliance-adjacent production systems: comfortable with on-call, incidents, regulator-driven requirements, and building the process scaffolding that prevents recurrence.
Care about measurement: you've built (or insisted on) the evals that prove a system does what it claims, and you've killed things that didn't.
Can drive ambiguous, multi-stakeholder tradeoffs (safety vs UX vs latency vs cost) to a decision and own the outcome.
Care deeply about AI safety and want your team's work to be the reason advanced models can be deployed at all.

Nice To Haves

Have worked in trust & safety, integrity, or abuse-prevention engineering at scale.
Experience with compliance-driven systems (child safety, copyright, age assurance) and the legal/policy interfaces they require.
Have shipped systems across multiple cloud providers and understand the parity/verification problems that creates.

Responsibilities

Hands-on lead and grow a team of engineers; own roadmap, OKRs, and execution.
Drive cross-functional work with ML Infra, Research, Product, Policy, and Legal - and with cloud partners for 3P deployment.
Set the bar for when an intervention is good enough to ship - backed by measurement - and represent safety and product tradeoffs to leadership and external stakeholders.
Own production reliability for intervention and compliance systems: incident response, postmortems, SLOs, and the verification processes that prevent repeat incidents.