AI Safety & Red Teaming

Location: San Francisco, CA

Work Model: Onsite

Industry: Applied AI / AI research data

Compensation: $130K-$150K base, $350K OTE

About the Company

Our partner is a fast-growing applied AI company that builds high-quality reinforcement-learning environments, agents, and datasets sold to the world's leading frontier AI labs. They recently grew from under twenty people to more than fifty in a matter of months and have reached a nine-figure revenue run rate. Quality is their core differentiator, and they are rapidly expanding into new domains.

The Opportunity

As the AI Safety & Red Teaming Lead, you will do hands-on adversarial work that directly shapes how foundation models behave. You will design the adversarial datasets and safety benchmarks that frontier labs use to improve their models, partnering directly with those labs on AI safety and model behavior.

This is a hands-on, IC-plus-lead seat for someone who has actually broken models, not just written about safety. You will join a small, high-agency team and help define how the company approaches safety work as it scales.

Responsibilities

• Design adversarial datasets and safety benchmarks used by frontier AI labs to improve foundation models.
• Jailbreak and stress-test frontier models to surface failure modes and unsafe behavior.
• Partner directly with lab researchers on model behavior, safety evaluation, and risk priorities.
• Write precise risk specifications grounded in harm taxonomies, safety frameworks, and model policy.
• Build automated red-teaming and evaluation pipelines in Python.
• Help set the standards and process for safety work as the function grows.

Requirements

• 2+ years in AI safety, red teaming, trust and safety, adversarial ML, or security research at a frontier AI lab, large tech company, top security firm, or equivalent.
• Hands-on adversarial experience such as jailbreaking frontier models, prompt-injection research, offensive security, penetration testing, bug bounty, CTFs, or trust and safety investigations.
• Fluency with harm taxonomies, evaluation design, safety frameworks, and model policy.
• Python proficiency and the ability to write production-quality code.
• Strong writing skills and the ability to specify risk precisely.
• High agency and a competitive track record.
• Bonus: published safety, alignment, or security research; bug bounty, CTF, or red-teaming competition record; depth in a high-consequence risk domain such as cyber, bio, fraud, child safety, or influence operations.
• Able to work onsite in San Francisco.