Applied AI & Machine Learning

Applied AI/ML Engineer, Security Agents

Build the models, datasets, evaluations, and learning systems that make autonomous security agents dependable in production.

Location: San Francisco Bay Area or Singapore
Employment: Full-time
Industry: Cybersecurity

FailSafe is building SWARM, an agentic offensive security platform that continuously tests software, validates exploitability, and helps security teams distinguish meaningful risk from noise.

As an Applied AI/ML Engineer, you will develop the data, models, evaluations, and production services behind SWARM's security agents. This role is for an applied builder who can move fluently between data science, machine learning, and software engineering, and who wants their models tested against reality, not just a benchmark.

You will work closely with senior engineers, security researchers, and product teams to turn real attack data into specialized models and reliable customer-facing capabilities. Your work may also support Project GlassBreak, our public-good responsible-disclosure program. This is a full-time role based in the San Francisco Bay Area or Singapore.

The opportunity

Why this role

Train and evaluate models against concrete security tasks, real attack evidence, and feedback from production rather than abstract proxy metrics alone.

Own meaningful work across the ML lifecycle, from dataset design and experimentation to inference services, monitoring, and product impact.

Develop rare depth at the intersection of applied AI and offensive security alongside senior engineers and security researchers.

What You’ll Do

Build data collection, cleaning, curation, labeling, and quality-control pipelines using findings from real-world attack surfaces and security research.
Train, fine-tune, and specialize small and large language models for detection, reasoning, classification, evidence extraction, and security adjudication tasks.
Design rigorous offline and online evaluations that measure accuracy, reliability, false-positive reduction, calibration, latency, cost, and customer impact.
Run disciplined experiments, analyze errors, identify model and data failure modes, and turn the results into concrete improvements.
Develop production ML services and APIs for inference, model orchestration, monitoring, versioning, and feedback collection.
Improve the multi-model quorum that reconciles outputs, adjudicates conflicting findings, and validates exploit paths within SWARM.
Work with security researchers to translate offensive security knowledge into datasets, task definitions, benchmarks, and model behavior.
Partner with software engineers to integrate model capabilities into scalable, observable, and secure customer-facing workflows.
Communicate findings clearly through experiment reports, evaluation dashboards, technical proposals, and production metrics.

Qualifications

Four to nine years of relevant professional experience across applied machine learning, data science, ML engineering, or software engineering.
Strong Python skills and experience with modern ML frameworks and ecosystems such as PyTorch, JAX, TensorFlow, scikit-learn, or Hugging Face.
Hands-on experience training, fine-tuning, evaluating, or deploying ML models, including LLMs, SLMs, embedding models, classifiers, or ranking systems.
Strong foundations in statistics, experimental design, data quality, evaluation methodology, and practical error analysis.
Experience building reliable data pipelines and working with SQL, structured and unstructured datasets, labeling systems, or feature stores.
Software engineering fundamentals including API design, testing, debugging, version control, code review, and production service ownership.
Familiarity with cloud platforms, containers, CI/CD, model serving, monitoring, and modern MLOps practices.
The ability to navigate ambiguous problems, learn unfamiliar technical domains quickly, and collaborate across research, engineering, security, and product.
Bachelor's degree or higher in computer science, machine learning, data science, engineering, mathematics, or a related field, or equivalent practical experience.

You’ll Be Especially Strong If You Have

Experience with supervised fine-tuning, preference optimization, reinforcement learning, synthetic data generation, or agent evaluation.
Built or operated LLM-powered or agentic systems in production, including their evaluation and observability layers.
Worked with cybersecurity data, vulnerability intelligence, source code, cloud telemetry, threat research, or other high-noise technical datasets.
Applied security knowledge across web applications, APIs, cloud infrastructure, secure coding, penetration testing, or red teaming.
Contributed to open-source ML tooling, published applied research, or built a strong portfolio of production ML systems.

Success Looks Like

Model and data improvements produce measurable gains on security tasks that matter to customers.
Evaluations reliably expose failure modes before models reach production and guide the team toward the highest-value work.
New research ideas move efficiently into observable, reliable, and maintainable product capabilities.
Security researchers and engineers can reuse the datasets, tooling, and ML infrastructure you build.
SWARM's agents become more accurate, explainable, and effective through your work.

Turn applied AI into security outcomes.

Train models on meaningful security problems, measure what works, and ship intelligence that helps defenders find and validate real risk.