Adversarial Machine Learning Engineer - Red Teaming

C-Serv Global Ltd; Canada; On-site; CAD 120,000 - 180,000; Full time

Benefits offered by this job

Fully remote Canada

Growth opportunities

Competitive compensation

Job summary

C-Serv Global Ltd. is seeking an Individual Contributor to join an existing client team focused on Guardrails and AI red-teaming for foundation models. You will tackle edge-case vulnerabilities, design and train ML models and Small Language Models in a security context.

You will work hands-on across model, data pipeline, and API layers, with a focus on reproducible findings, remediation guidance, and retesting. Join a values-driven, fully remote team across Canada.

Qualifications

• Proficient in Python with ML frameworks (PyTorch, TensorFlow, Hugging Face).
• Experience designing and executing adversarial attacks and defenses.
• Strong threat-modeling and security evaluation skills.

Responsibilities

• Hands-on adversarial testing across model, application, and data pipeline.
• Translate findings into actionable remediation steps and retest.
• Collaborate with client's Guardrails/AI red-teaming team and deliver clear risk communications.
• Design and train ML/SLM models to improve robustness.
• Maintain up-to-date knowledge of adversarial ML and GenAI security research.

Skills

Python

ML frameworks

LoRA/QLoRA

Adversarial ML

Threat modeling

ML security

Explainable AI

Communication

Tools

ART (Adversarial Robustness Toolbox)

CleverHans

Foolbox

Job description

We are looking for that Individual contributor, who will slot into an existing client team already running Guardrails and AI red teaming for their foundation model suite. The role digs into edge-case vulnerabilities that campaign reports surface but don't fully explain, the ideal candidate will be able to design and train ML models as well as SLM's in the security context.

What You’ll Own

• Hands-on adversarial testing across the model, the application and agentic layer, and the data pipeline: multi-turn jailbreaks and guardrail bypass, prompt injection, agent and tool-chain misuse, dangerous-capability evaluation, API abuse, and, where relevant, data poisoning, model inversion and membership inference.
• Digging deeper into edge-case findings from AI red-team campaigns, turning a flagged anomaly into a fully understood, reproducible vulnerability.
• Severity-ranked findings mapped to the OWASP Top 10 for LLM Applications, the NIST AI Risk Management Framework and its Generative AI Profile, MITRE ATLAS, and EU AI Act Article 55 expectations, with evidence and clean reproduction steps.
• Remediation guidance that's actually usable, and a retest to confirm the fixes hold.

The Human Side of It

The testing is the craft. The trust is the job. Findings only matter if the right people understand and act on them.

• Works shoulder to shoulder with the client's Guardrails and AI red-teaming team, not at a distance from them.
• Translates findings into plain language: technical depth for the engineers, a clear risk picture for anyone less hands-on with the model itself.
• Stays embedded well past the first findings, through remediation, to the retest that proves it's fixed.

• Expert-level Python programming with deep proficiency in ML frameworks such as PyTorch, TensorFlow, and Hugging Face Transformers
• Hands-on experience fine-tuning ML models and Small Language Models (SLMs) - including techniques such as LoRA/QLoRA, PEFT, instruction tuning, and domain adaptation - for both performance and robustness objectives
• Strong foundation in ML mathematics: optimization, linear algebra, probability, and statistics
• Proven ability to design and execute adversarial attacks, including evasion (adversarial examples), data poisoning, model extraction, and membership inference
• Experience implementing defenses such as adversarial training, robust fine-tuning, input sanitization, and differential privacy
• Proficiency with adversarial ML toolkits such as Adversarial Robustness Toolbox (ART), CleverHans, and Foolbox
• Experience red-teaming AI/LLM systems, including prompt injection, jailbreak testing, and safety/alignment evaluation
• Ability to evaluate and benchmark model robustness, safety, and security posture before and after fine-tuning
• Familiarity with MLOps practices - model versioning, experiment tracking, and secure deployment pipelines
• Strong threat-modeling skills and an attacker's mindset, with the ability to communicate risks clearly to technical and non-technical stakeholders
• Active awareness of the latest adversarial ML and GenAI security research

Fully remote working anywhere in the Canada, built around delivery rather than presence.

A clear path to grow into staff and principal-level technical influence.

Full support from C-Serv across the hiring process and beyond, with full-cycle accountability.

A values-led, woman-owned delivery partner built on empathy, integrity, collaboration, and growth.