Senior AI Red Team Researcher

Hibari Tech

Remote · $140K – $170K · Full-time · Posted 6 months ago

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

This role is no longer accepting applications

Don't miss the next role like this

Create a free profile with your skills. We match you to new AI roles as they open, and companies hiring can contact you directly.

Create free profile, 2 minutes

Similar open roles you can apply to now:

See all open AI jobs

About the role

Our team exists to break AI systems before adversaries do. We are a cybersecurity firm's AI research division, and our work falls into two categories: red teaming ML systems deployed by our clients — finding the ways they fail under adversarial conditions — and building the detection and hardening tooling that makes those failures less likely or less damaging. This is not a conventional ML role. The question we ask every day is not "how do we maximise accuracy?" but "how would a motivated adversary make this model fail, and what does that failure look like when it happens at scale?" We work on adversarial examples, model inversion, membership inference, prompt injection in LLM deployments, and data poisoning scenarios. If those concepts are familiar to you from a defensive or offensive research context, we'd like to talk. The team is small — six researchers and two engineers. We publish occasionally when the work is complete enough and when publication doesn't compromise client engagements. We engage with the research community at venues like IEEE S&P and USENIX Security. We are funded and serious.

Fully remote.

Responsibilities

Conduct adversarial red team evaluations of client ML systems and produce written reports with severity ratings and remediation recommendations
Research and implement novel attack techniques against production ML systems
Build tooling to automate components of the red team workflow for common attack classes
Contribute to our internal knowledge base of attack patterns, model architectures, and mitigations
Present findings to client security teams and, where appropriate, contribute to publications and conference presentations

Requirements

5+ years of ML engineering or research, with a demonstrable focus on adversarial robustness, security, or safety
PyTorch — you implement attacks from scratch, not just run existing libraries against pre-trained checkpoints
Deep understanding of adversarial ML: FGSM, PGD, and more recent threat models for both vision and language systems
Python for attack implementation, evaluation pipelines, and automation of red team workflows
AWS for deploying, testing, and isolating adversarial evaluation environments
Docker for reproducible attack environments and client-deployable tooling
Bonus: experience with LLM-specific vulnerabilities — jailbreaking, prompt injection, and indirect prompt injection in agentic systems

Benefits

Work on the most interesting failure modes in deployed ML — real systems, real adversaries, real consequences
Full remote with occasional client travel (covered)
$140,000 – $170,000 base salary + annual bonus
$3,000 annual conference and research budget
Publication support — time and resources to write and present significant findings