Job description

Full-time | Remote | Adversarial ML | Reports to Head of Risk About Elloe Elloe is the trust layer for AI. We sit between the world’s most powerful language models and the institutions that can't afford to get it wrong — hospitals, banks, regulators. We trace and block failures in real time. That’s not marketing — we’re deployed at the European Commission, with NIH clinical trials, and inside a Top-5 EU bank catching GDPR violations live. This is the enforcement layer GenAI has been missing. We're not visualizing problems — we're fixing them. About the Role Elloe’s safety loop only works if we can simulate attacks before they reach real users. You’ll drive red teaming at the system level — from jailbreaking to fuzzing and help shape the defense logic in AutoHeal and ReplayHeatmap. What You’ll Build 1. Red Team Simulation Engine • Launch adversarial attacks (prompt injections, bypass chains, logic traps) • Generate incident traces used to train AutoHeal patch logic • Contribute to fuzzing harnesses, risk scoring, and breach labeling 2. Security Patch Forecasting • Forecast deployment risks based on incident graph trends • Automate “patch windows” based on SHAP mismatch clusters • Help convert audit traces into product-stoppable violations 3. System-Level Defense • Build explainability-linked guardrails across the stack • Collaborate with infra and explainability leads on enforcement crossover Who You Are • Deep experience in adversarial ML, red teaming, or fuzz testing • Understands how explainability can be a security surface • Bonus: experience with threat modeling, diff-testing, or jailbreak detection Why It Matters Red teaming isn’t an afterthought. It’s how Elloe gets trusted to run in places that can’t afford to fail. Why Now Major regulators are asking for test results before they approve AI deployment. Institutions want defensible logs, not theoretical attacks. This is the moment when red teaming goes from research to required. You’ll Leave This Role With • A safety portfolio tied to real red team incidents, not just demos • Impact on critical infrastructure used by hospitals, central banks, and governments • A seat at the table defining what trustworthy GenAI actually means Logistics & Application • Start Date: Flexible (Q3 ideal) • Location: Remote-first; timezone overlap with NY or DC preferred • Comp: Competitive salary + equity • To Apply: Share one failure mode you’d simulate in a next-gen red team harness.

Requirements

Must-haves

Full-time | Remote | Adversarial ML | Reports to Head of Risk
What You’ll Build
Red Team Simulation Engine
Launch adversarial attacks (prompt injections, bypass chains, logic traps)
Generate incident traces used to train AutoHeal patch logic
Contribute to fuzzing harnesses, risk scoring, and breach labeling
Security Patch Forecasting
Forecast deployment risks based on incident graph trends
Automate “patch windows” based on SHAP mismatch clusters
Help convert audit traces into product-stoppable violations
System-Level Defense
Build explainability-linked guardrails across the stack
Collaborate with infra and explainability leads on enforcement crossover
Deep experience in adversarial ML, red teaming, or fuzz testing
Understands how explainability can be a security surface
Why It Matters
Red teaming isn’t an afterthought. It’s how Elloe gets trusted to run in places that can’t afford to fail.
Major regulators are asking for test results before they approve AI deployment. Institutions want defensible logs, not theoretical attacks. This is the moment when red teaming goes from research to required.
You’ll Leave This Role With
A safety portfolio tied to real red team incidents, not just demos
Impact on critical infrastructure used by hospitals, central banks, and governments
A seat at the table defining what trustworthy GenAI actually means

Nice-to-haves

Bonus: experience with threat modeling, diff-testing, or jailbreak detection
Location: Remote-first; timezone overlap with NY or DC preferred

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

mlai

Platform: gem
Effort: medium