Anthropic
Staff+ Software Engineer, Safeguards
$320K–$485K
San Francisco, CA +3 · Staff · Full-time · Posted Oct 14, 2025
- Location
- San Francisco, CA
New York City, NY
New York, New York, United States
San Francisco, California, United States - Experience
- Years required correlate with internal job level; strong candidates may have 8+ years of software engineering experience.
- Education
- Bachelor’s degree or an equivalent combination of education, training, and/or experience; relevant field of study demonstrated through coursework, training, or professional experience.
- Visa sponsorship
- Visa sponsorship is available, but not guaranteed for every role or candidate; the company will make every reasonable effort to obtain a visa after an offer.
About the job
Build safety and oversight systems that monitor AI models, detect unwanted behavior and abuse, and prevent disallowed use.
What you'll do
- Develop monitoring systems and dashboards to identify unwanted behavior and support automated enforcement or analyst review.
- Build abuse detection mechanisms and scalable, multi-layered real-time safety defenses.
- Identify abuse patterns and share them with research teams to improve model training.
What they ask for
- Bachelor’s degree in Computer Science, Software Engineering, or comparable professional experience.
- Proficiency in at least one programming language, such as Python or Java.
- Ability to work across the stack.
- Strong communication skills, including explaining technical concepts to non-technical stakeholders.
Nice to have
- Experience with integrity, spam, fraud, or abuse detection and mitigation; trust and safety mechanisms for AI/ML systems.
- Experience with prompt engineering, jailbreak attacks, adversarial inputs, or internal tooling for operational teams.
Pay
$320,000 — $485,000 USD annual base salary
Tools
Python, Java
AI focus
Build monitoring, abuse detection, and enforcement systems for AI models; surface patterns to researchers to harden models during training.
Model-written summary · the original posting is authoritative