Hybrid Software Engineer — AI Safety Evaluations

Anthropic; San Francisco (CA); Hybrid; USD 320,000 - 485,000; Full time

Job summary

Anthropic is seeking a skilled engineer to build evaluation infrastructure for detecting AI misuse. In this role, you'll craft experiments and datasets to enhance the reliability of Claude's safety systems.

Ideal candidates should have a strong Python background, experience with data pipelines, and a comprehensive understanding of LLM capabilities and their implications in safety mechanisms.

The role emphasizes trust in automated abuse detection methods, contributing to AI's beneficial development.

Qualifications

• 6+ years of industry software engineering experience preferred.
• Experience in trust and safety or abuse detection systems is a plus.
• Strong communication skills and an understanding of AI system risks.

Responsibilities

• Own the evaluation harness for an investigation system.
• Construct high-quality eval datasets representing real-world misuse.
• Measure agent performance and identify measurement gaps.
• Productionize research for agent changes and model upgrades.
• Build tools for policy experts to author evaluations.

Skills

Proficiency in Python

Experience in building data pipelines

Working understanding of LLMs

Strong data analysis skills

Ability to translate ambiguous problems

Education

Bachelor’s degree or equivalent

Tools

Large-scale data processing

Job description

Anthropic is seeking a skilled engineer to build evaluation infrastructure for detecting AI misuse. In this role, you'll craft experiments and datasets to enhance the reliability of Claude's safety systems.

Ideal candidates should have a strong Python background, experience with data pipelines, and a comprehensive understanding of LLM capabilities and their implications in safety mechanisms.

The role emphasizes trust in automated abuse detection methods, contributing to AI's beneficial development.