Work on problems with real consequences
We're a small team building the adversarial test bench for the AI that runs the modern world. If you want your work to matter, and to be measured, you'll fit here.
What we optimize for
Attack to defend
We break things on purpose so they hold when it counts. Offense is how we earn the right to advise on defense.
Evidence over assertion
Robustness is measured, not claimed. Every result we ship is one we could defend in front of the people who broke it.
Consequences are the point
We work on models that steer real systems. The stakes are why the work matters, and why we hold a high bar.
Where we're hiring
High-ownership, remote-first in the US, with time together when the work calls for it.
Adversarial ML Research Scientist
Push the state of the art in adversarial attacks and defenses for perception and decision models, and turn that research into capabilities that ship in the platform. You'll work across gradient, physical, and black-box attacks, and the defenses that answer them.
- Design and implement novel attacks against vision, sensor, and multimodal models
- Advance physical-adversarial methods (EOT, patches, camouflage) toward operational realism
- Evaluate and harden defenses, adversarial training, certified smoothing, detection
- Translate research into production attack modules and reproducible benchmarks
- PhD or equivalent research experience in ML, computer vision, or security
- Publications or demonstrable work in adversarial ML or a closely related area
- Fluency in PyTorch and the modern attack/defense literature
- The instinct to break your own results before anyone else does
Senior ML Platform Engineer
Own the engine room: the run orchestration, attack pipelines, and pluggable model backends that make adversarial evaluation fast, reproducible, and safe to run against real models. This is the core of the product.
- Build and scale the attack/run orchestration across vision, audio, text, and signal
- Integrate and maintain pluggable backends (IBM ART / HEART) behind a clean abstraction
- Make evaluation reproducible, observable, and resource-safe under load
- Ship the autonomous ONNX introspection-and-test pipeline forward
- Strong Python and systems fundamentals; comfortable in PyTorch and ONNX Runtime
- Experience building ML infrastructure or evaluation tooling in production
- A bias for correctness and reproducibility over cleverness
- Bonus: exposure to adversarial ML, detection models, or MLOps at scale
Founding Full-Stack Engineer
Own the surface customers actually touch, the interactive platform, reports, and the marketing site, end to end. Next.js and React on the front, Python services behind. You'll set patterns the rest of the team builds on.
- Build the interactive demo, run dashboards, and reporting experiences
- Design clean, fast interfaces for genuinely complex ML output
- Connect the frontend to Python evaluation services through well-shaped APIs
- Sweat the details that make the product feel serious and trustworthy
- Strong TypeScript / React / Next.js and an eye for interface quality
- Comfortable working across the stack into Python services
- Product sense, you can turn a vague need into a shipped, considered feature
- Startup-ready: high ownership, low ceremony
Offensive ML Security Engineer
Run real adversarial engagements against customer models and infrastructure. Half red-team craft, half ML, you'll find the failure, prove it cleanly, and hand back a report a security team can act on.
- Scope and execute adversarial assessments of deployed CV/NLP models
- Chain model-level attacks with deployment and pipeline weaknesses
- Produce rigorous, reproducible findings mapped to MITRE ATLAS
- Advise customers on hardening and coordinated disclosure
- Security engineering or red-team background, with real ML depth
- Hands-on with adversarial attacks and the tooling around them
- Clear technical writing, a finding is only as good as its report
- Comfort operating in air-gapped and regulated environments
Forward-Deployed Engineer
Be the technical point of contact for the teams adopting Mirage. You'll stand up evaluations against their models, interpret the results with them, and carry hard-won field lessons back into the product.
- Onboard customers and run evaluations against their real models
- Translate robustness findings into concrete engineering next steps
- Build integrations and glue that unblock deployments
- Feed the sharpest edges from the field back to Platform and Research
- Engineer who is equally comfortable in a terminal and a customer meeting
- Solid Python and ML literacy; can reason about a model's failure modes
- Calm under ambiguity and travel; strong communicator
- Bonus: experience in regulated, high-stakes, or on-prem environments
Don't see your role?
If you're exceptional at something adjacent to this work, tell us what you'd build. We hire for talent and trajectory, not just open reqs.