Lead Adversarial Engineer
- Role
- Security
- Experience
- Lead
- Employment
- Full-time
Open to US only. Set where you work from to check your eligibility.
No BS summary
Lead adversarial/red-team engineer for frontier AI models. Needs experience planning and running jailbreak, prompt injection, and tool-abuse campaigns, plus strong experimentation and stakeholder reporting skills. US remote only.
ABOUT US
Our mission is to raise AGI with the richness of human intelligence — curious, witty, imaginative, and full of unexpected brilliance.
Surge was founded by engineers and researchers who dreamed of building the next generation AI. We're building a platform that powers the most powerful models in the world in partnership with companies like Anthropic, Google, Microsoft, and Meta.
At Surge, we believe the path to AGI isn't just about scaling compute—it's about embracing the unlimited ceiling of human intelligence and creativity in the data that shapes these systems. Our platform combines elite human expertise with cutting-edge tools for scalable oversight, from building rich RL environments to conducting rigorous evaluations that go beyond benchmarks. We've run a profitable business from day one without raising venture funding.
THE ROLE
As a Lead Adversarial Engineer, you’ll run end-to-end red-teaming workstreams against frontier models — scoping threat models, designing campaigns, coordinating operators, and synthesizing results into clear risk pictures and decision-ready reports. You’ll orchestrate structured adversarial exercises across modalities and tools, ensuring coverage, reproducibility, and crisp learning loops.
You won’t just find failures — you’ll build the operational engine that repeatedly surfaces them under realistic constraints. This is a role for someone who thrives on program design, loves turning messy attack spaces into disciplined test plans, and can drive cross-functional execution from kickoff to readout.
WHAT YOU'LL DO
- Stand up a recurring red-team cadence: scoping targets, recruiting operators, defining success criteria, and executing multi-week campaigns
- Create scenario banks and attack taxonomies; ensuring breadth/depth coverage and tracking families of exploits across versions and contexts
- Produce executive readouts and issue trackers that distill severity, exploitability, and user harm, with crisp reproduction steps and artifacts
- Partner with research, product, and ops teams to validate fixes and rerun focused regressions; maintaining dashboards for trendlines and residual risk
WHAT WE’RE LOOKING FOR
- Red-Team Program Leadership – Experience planning and running adversarial campaigns (jailbreaks, prompt injection, tool abuse), including playbooks, ops cadence, and after-action reviews
- Methodical Experimentation – Strength in designing scenarios, controls, and metrics; comfort triaging findings and prioritizing next passes based on evidence
- Stakeholder Command – Ability to brief partners, align on objectives, and translate results into actionable remediation tracks with clear owners and timelines
What you'll do
- Run end-to-end red-teaming workstreams against frontier models.
- Scope threat models.
- Design campaigns.
- Coordinate operators.
- Synthesize results into clear risk pictures and decision-ready reports.
- Orchestrate structured adversarial exercises across modalities and tools.
- Ensure coverage, reproducibility, and crisp learning loops.
- Build the operational engine that repeatedly surfaces model failures under realistic constraints.
- Stand up a recurring red-team cadence: scoping targets, recruiting operators, defining success criteria, and executing multi-week campaigns.
- Create scenario banks and attack taxonomies.
- Ensure breadth and depth coverage.
- Track families of exploits across versions and contexts.
- Produce executive readouts and issue trackers that distill severity, exploitability, and user harm, with crisp reproduction steps and artifacts.
- Partner with research, product, and ops teams to validate fixes and rerun focused regressions.
- Maintain dashboards for trendlines and residual risk.
What they require
- Experience planning and running adversarial campaigns, including jailbreaks, prompt injection, and tool abuse.
- Experience creating playbooks, ops cadence, and after-action reviews for red-team programs.
- Strength in designing scenarios, controls, and metrics.
- Comfort triaging findings and prioritizing next passes based on evidence.
- Ability to brief partners, align on objectives, and translate results into actionable remediation tracks with clear owners and timelines.
- Thrives on program design.
- Loves turning messy attack spaces into disciplined test plans.
- Can drive cross-functional execution from kickoff to readout.
Surge is building a platform that powers AI models in partnership with companies like Anthropic, Google, Microsoft, and Meta, combining human expertise with tools for scalable oversight, RL environments, and evaluations.