AI Security Scientist
Advance the applied research behind Hardshell's detection and transformation methods, from measurement methodology to new defenses. Publication-quality rigor under production constraints.
The role
Hardshell's product is built on measurement: how much a given AI system can leak, whether a pattern of queries is extraction, and whether a hardening step kept the answers people need. You'll work alongside our founding AI security scientist and engineer on the methods behind those answers, and see your work run against real customer systems.
What you'll do
- Design and run experiments on how sensitive data leaks through AI systems, including extraction from AI knowledge bases, training data memorization, membership inference, and embedding inversion.
- Develop and validate methods that detect extraction in retrieval traffic, including attacks spread across many requests, with measured precision and false-positive rates.
- Advance our dataset hardening methods and quantify the tradeoff between what an attacker can recover and what the system can still answer.
- Build the measurement methodology behind our exposure scores so every finding holds up to review by a customer's security team.
- Work with engineering to take methods from notebook to production, and contribute to publications, technical guides, and federal research proposals.
What you bring
- An MS or PhD in computer science, statistics, machine learning, or a related field, or equivalent research experience.
- Research experience in adversarial ML, privacy-preserving ML, ML security, or applied statistics, with a record of rigorous work in publications, open source, or equivalent.
- Strong Python and experience with modern ML tooling and with LLM and embedding models.
- Sound statistical judgment in experimental design, evaluation, and deciding when a result is real.
- The ability to write clearly about methods and results for technical and non-technical readers.
- U.S. citizenship.
Nice to have
- Background in differential privacy, anomaly detection, or information retrieval.
- Experience with RAG systems or vector search in production.
- Prior work on federally funded research, such as SBIR/STTR or DoD and intelligence community programs.
The details
- Location: Hybrid in Charlottesville, VA, or remote on Eastern Time.
- Compensation: Up to $160K base salary, plus equity and health coverage.
- Eligibility: U.S. citizenship is required for this role because of our federal work.
About Hardshell
Hardshell secures the data layer of enterprise AI. Our platform measures what an AI deployment can leak, catches extraction from AI knowledge bases as it happens, and hardens sensitive data before any model touches it. It is model-agnostic and runs in the customer's environment, so sensitive data stays where it lives.
We're a seed-stage team headquartered in Charlottesville, Virginia. Our founders come from national security AI and cyber operations: our CTO spent five years advising the Department of Defense on AI security at the Joint AI Center and the CDAO, and our CEO served as an Army Cyber officer with assignments across U.S. Cyber Command and USSOCOM. We work with commercial customers in regulated industries and with federal and national security partners.
How to apply
Apply through the form on our careers page with your resume and LinkedIn profile. If more than one role fits, apply to the closest and say so in your note. Questions can go to careers@hardshell.ai.