Director Offensive Security Engineer Jobs in California

About the role:

Cobalt is seeking experienced offensive security practitioners to contribute expert reasoning, adversarial testing, and evaluation data used to train and assess frontier AI models.

This opportunity is suited to people who have worked as penetration testers, red team operators, security researchers, vulnerability researchers, exploit developers, application security engineers, or bug bounty hunters, in consultancies, internal security teams, or independently. Both traditional offensive security experience and experience probing AI systems are relevant, and you do not need both.

You do not need prior experience in data annotation or AI research. You must, however, be able to find and reason about real weaknesses in software or in model behavior unaided, and you must be comfortable documenting your approach clearly in writing.

All work is performed against sandboxed environments, purpose-built targets, and model endpoints supplied by us or by the lab. We do not accept work performed against systems you are not authorized to test, and we do not accept material obtained without authorization.

What you'll do:

Depending on the project, you may:

• Produce written attack traces on security tasks, capturing how you form and test hypotheses, what you rule out and why, and how you arrive at a working approach, rather than only the end result
• Adversarially test model behavior, probing for prompt injection, unsafe tool use, data exfiltration paths, and failures of refusal or policy adherence, and document reproducible cases
• Author novel security problems, capture-the-flag style challenges, and evaluation scenarios with verifiable success criteria
• Evaluate model-generated security content and code: rank responses, explain what makes the stronger one stronger, and identify the specific step at which the reasoning or the exploit logic breaks down
• Design rubrics and partial-credit criteria for scoring multistep offensive and defensive tasks

Projects follow their own guidelines, scope rules, and quality standards, and you will work with feedback from reviewers and lab research teams.

Required qualifications:

• Demonstrable offensive security experience, evidenced by professional penetration testing or red team engagements, published vulnerability research or CVEs, a substantive bug bounty record, competitive CTF results, or comparable work
• Strong hands-on coding ability in at least one of Python, C, C++, Go, Rust, or JavaScript, sufficient to read unfamiliar codebases and write your own tooling
• Depth in at least one area, for example web and API security, cloud and container security, network and infrastructure testing, binary exploitation and reverse engineering, or AI and LLM security
• Ability to explain each step of your reasoning clearly in writing, and to produce documentation another practitioner could reproduce
• Willingness to work strictly within defined scope and authorization, and to sign a confidentiality agreement covering project materials

Certifications such as OSCP, OSWE, OSEP, GPEN, or GXPN are useful but not required, as is prior experience with AI red teaming, model evaluation, or safety research.

Why Join Cobalt AI:

• Advance frontier AI where it counts. Apply your offensive expertise to data that frontier labs cannot obtain any other way, where your judgment directly shapes how the next generation of models handles security reasoning and resists misuse.
• Grow professionally. Expand your influence through evaluation projects, advisory roles, and research collaborations, while developing a working understanding of how frontier models are trained and assessed.
• Work with a top-tier network. Collaborate with security researchers and practitioners from leading organizations on high-impact, flexible work.
• Set your own schedule. Flexible 10 to 40 hour weeks that fit around your existing engagements and your life.
• Competitive pay. Rates vary by project and are determined by a number of factors, including scope, skillset, and experience.