Staff+ Software Engineer, Safeguards Evals

Details

Anthropic is building reliable, interpretable, and steerable AI systems. This role sits at the intersection of applied machine learning research and software engineering, developing the methods and infrastructure used to evaluate model behavior and agentic systems for safety, misuse, prompt injection, and user well-being.

Responsibilities

• Design and run experiments to improve evaluation quality, including methods for generating representative test data, simulating realistic user behavior, and validating grading accuracy.
• Research how multi-turn conversations, tools, long context, and user diversity affect model safety behavior.
• Analyze evaluation coverage, identify measurement gaps, and evolve evaluations so they remain high-signal as model and agent capabilities advance.
• Build and own the evaluation harness for an agentic investigation system, including metrics, test cases, and grading approaches for a complex, long-horizon agent.
• Measure agent performance end-to-end, including detection precision and recall, investigation quality, and robustness.
• Drive hill-climbing on difficult harm areas and construct reinforcement learning environments to improve Claude’s safety investigation capabilities.
• Build high-quality evaluation datasets representing real-world misuse across areas such as cyber attacks, biological weapons, and influence operations.
• Draw on real traffic patterns and synthetic generation, and collaborate with Policy and Enforcement to translate observed harm patterns into measurable evaluations.
• Productionize successful research in evaluation, regression, and release pipelines that run during model training, agent changes, prompt updates, underlying model upgrades, and after launch.
• Build tooling that enables policy experts to author, run, and iterate on evaluations without engineering support.
• Surface findings to research and training teams to drive upstream model improvements.

Requirements

• Experience building and maintaining data pipelines.
• Experience working with large language models and a working understanding of their capabilities and failure modes, especially agentic systems with tool use and multi-step reasoning.
• Strong data analysis skills and the ability to draw reliable insights from large datasets.
• Ability to move between research prototyping and production-quality code.
• Ability to translate ambiguous problems into concrete, testable experiments.
• Strong interest in AI safety and its real-world impact.
• Preferred: 8 or more years of industry software engineering or machine learning engineering experience.
• Preferred: Expertise in LLM or agent evaluation frameworks, benchmarks, or automated grading systems.
• Preferred: Extensive experience in trust and safety, content moderation, or abuse detection systems.
• Preferred: Experience in red teaming, adversarial testing, or jailbreak research on AI systems.
• Preferred: Experience with synthetic data generation or data augmentation.
• Preferred: Experience with distributed systems or large-scale data processing.
• Preferred: Experience with prompt engineering or building LLM-powered applications.
• Minimum education is a bachelor’s degree or equivalent combination of education, training, and experience. The field of study must be relevant to the role as demonstrated through coursework, training, or professional experience.

Compensation

The annual salary range is $320,000–$485,000 USD.

Benefits and Logistics

• Hybrid policy: Staff are currently expected to work from one of Anthropic’s offices at least 25% of the time, although some roles may require more office time.
• Anthropic sponsors visas for some roles and candidates and will make every reasonable effort to obtain a visa when an offer is made.
• Anthropic offers competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, flexible working hours, and office collaboration spaces.