Technical Program Manager, Safeguards (Infrastructure & Evals)

Details

Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. The Safeguards Engineering team builds and operates infrastructure that keeps Anthropic's AI systems safe in production — classifiers, detection pipelines, evaluation platforms, and monitoring systems that sit between models and the real world. This role focuses on operational health and forward momentum for that stack, driving reliability through incident response, SLO definition and tracking, runbook quality, and cross-team platform projects.

Responsibilities

• Own the Safeguards Engineering ops review: run recurring cadences to surface incidents, reliability trends, and ensure the right participants and decisions.
• Drive incident tracking and post-mortem execution: track incidents across the organization (including partner teams like Inference), ensure post-mortems are written, and close out action items.
• Establish and maintain SLOs with partner teams: define service-level objectives for safety-critical pipelines and build tracking/reporting to measure them.
• Maintain runbook quality and incident-ownership clarity: keep runbooks accurate and ensure incident ownership is unambiguous for safety-critical areas (e.g., account-banning false positives, CSAM detection).
• Drive platform migrations and infrastructure projects: program-manage migrations (platform, incident platform, cloud monitoring systems) and other cross-team infrastructure work so systems remain operational during transitions.
• Coordinate evals platform improvements: partner with the evals engineering team to scope work, track dependencies, and drive improvements to evaluation and self-serve capabilities.

Requirements

• Solid technical program management experience, especially in operational or infrastructure-heavy environments where you own both ongoing operational cadence and discrete projects.
• Strong understanding of how production ML systems operate to triage incidents and have technical conversations with engineers (you don't need to write the code, but must follow the technical thread).
• Demonstrated ability to close loops: track and drive post-mortem actions, ensure SLOs are checked, and keep runbooks up to date.
• Ability to work effectively across team boundaries and coordinate with partner teams where you don't have direct authority.
• Comfortable context-switching between "keep the lights on" work and longer-horizon platform projects.
• Experience with or strong interest in AI safety and appreciation for the distinct reliability needs of safety-critical pipelines.

Strongly preferred / helpful experience

• Experience with SRE practices, incident management frameworks, or on-call operations at scale.
• Experience with evaluation infrastructure for ML systems and understanding of how evals are designed, run, and interpreted.
• Experience driving infrastructure migrations in complex, multi-team environments without taking operational systems offline.
• Familiarity with monitoring and alerting tooling and operational culture (PagerDuty, Datadog, or equivalents).

Logistics

• Education: At least a Bachelor's degree in a related field or equivalent experience.
• Location-based hybrid policy: staff are expected to be in one of Anthropic's offices at least 25% of the time (some roles may require more time in-office).
• Visa sponsorship: Anthropic states they do sponsor visas and retain an immigration lawyer to assist, though sponsorship is not guaranteed for every role/candidate.
• Deadline to apply: None (applications received on a rolling basis).

Compensation and Benefits

• Annual salary range: $290,000 - $365,000 USD.
• Anthropic offers competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, flexible working hours, and office space to collaborate.

Additional notes

• Work involves close coordination with partner teams like Inference and Cloud Inference, and responsibilities include incident ownership for safety-sensitive areas such as CSAM detection and account-banning false positives.
• Candidates are encouraged to apply even if they do not meet every qualification.