Anthropic
Cyber Evaluations Engineer
$300K–$405K
Remote-Friendly, United States +3 · Remote · Full-time · Posted Sep 1, 2026
- Location
- Remote-Friendly, United States
San Francisco, CA
Washington, DC
San Francisco, California, United States - Experience
- Years of experience required will correlate with the internal job level requirements for the position.
- 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 for every role or candidate; if an offer is made, Anthropic will make every reasonable effort to obtain a visa.
About the job
Build and run evaluations of cyber capabilities and safeguard robustness in AI models, and develop probes and detection systems for cyber misuse.
What you'll do
- Design and run cyber capability, uplift, and safety evaluations; perform robustness testing before major releases.
- Analyze evaluation results and communicate findings to teams and stakeholders.
- Build evaluation tooling and cyber-misuse detection probes, collaborating with policy and engineering teams on safeguards.
What they ask for
- Experience building or running evaluations, benchmarks, or test suites for software or ML systems, including delivering results on short, fixed timelines.
- Hands-on cybersecurity experience, such as CTF participation, vulnerability research, exploit development, or security research.
- Proficiency in Python.
- Strong ability to communicate evaluation results with cross-functional or policy stakeholders.
Nice to have
- Deep offensive-security or security-research experience, including building AI security benchmarks.
- Experience analyzing adversarial or abuse data, AI/ML evaluation frameworks, or detection content.
Pay
$300,000 — $405,000 USD (base salary)
Tools
Python, Sigma, YARA, Suricata, SIEM
AI focus
Evaluate cyber-relevant capabilities and safeguard robustness in models; develop abuse-detection probes and architecture.
Model-written summary · the original posting is authoritative