Cyber Evaluations Engineer

We’re hiring Cyber Evaluations Engineers to build and run the evaluations that measure cyber-relevant capabilities and safeguard robustness in our models. You’ll design new evals, run per-release robustness testing, and dig into data on jailbreaks and prompt bypasses to understand where our safeguards hold up and where they don’t. You’ll also design many of the probes that detect cyber abuse in production and help shape the overall detection architecture alongside the policy team., Design and run capability, uplift, and safety evaluations to assess cyber-relevant risk in new models

Execute per-release safeguard-robustness testing ahead of major model launches

Analyze evaluation results and communicate findings clearly to the team and to stakeholders

Design, prototype, and tune detection probes for cyber misuse

Work with the cyber policy team to turn policy lines into a layered, robust abuse-detection architecture, and measure its precision and coverage over time

Build and maintain internal tooling used to run and score evaluations

Collaborate with policy and engineering partners to translate eval findings into safeguard improvements

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 (e.g., CTF participation, vulnerability research, exploit development, or security research)

Proficiency in Python

Strong ability to communicate evaluation results with multiple cross-functional stakeholders or potential policy stakeholders, Deep offensive-security or security-research experience, including experience building AI security benchmarks

Experience analyzing adversarial or abuse data (e.g., jailbreaks, prompt bypasses, intrusion or fraud telemetry)

Experience working onsite with government partners on testing or evaluation engagements

Experience with AI/ML evaluation frameworks

Familiarity with coordinated vulnerability disclosure practices

Experience testing pre-release or pre-deployment software or models under confidentiality constraints

Experience authoring detection content (e.g., Sigma, YARA, Suricata, SIEM rules) or building ML-based abuse detection

Active secret security clearance or higher, or eligibility to obtain one, Minimum education: Bachelor’s degree or an equivalent combination of education, training, and/or experience

Required field of study:A field relevant to the role as demonstrated through coursework, training, or professional experience

Minimum years of experience: Years of experience required will correlate with the internal job level requirements for the position

Location-based hybrid policy: Currently, we expect all staff to be in one of our offices at least 25% of the time. However, some roles may require more time in our offices.

Anthropic is a public benefit corporation headquartered in San Francisco. We offer competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, flexible working hours, and a lovely office space in which to collaborate with colleagues. Guidance on Candidates’ AI Usage: Learn aboutour policy for using AI in our application process. #J-18808-Ljbffr

We believe that the highest-impact AI research will be big science. At Anthropic we work as a single cohesive team on just a few large-scale research efforts. And we value impact - advancing our long-term goals of steerable, trustworthy AI - rather than work on smaller and more specific puzzles. We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science. We’re an extremely collaborative group, and we host frequent research discussions to ensure that we are pursuing the highest-impact work at any given time. As such, we greatly value communication skills.

The easiest way to understand our research directions is to read our recent research. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.