Frontier AI Security Researcher

Build something people love

We’re building a new AI-enabled adversarial testing capability whose mandate is simple but ambitious: find all the ways Wealthsimple can be exploited before our AI-enabled adversaries do. This group combines penetration testing, secure code analysis, and attack simulation R&D to continuously probe Wealthsimple’s systems using a combination of automation, autonomous AI agents, and human expertise. If you want to build an automated, end-to-end clearbox pentesting/red teaming platform, this is the team doing it!

You’ll join as an individual contributor on this team, reporting into the Manager, Application Security, working alongside one or more threat hunters/adversarial simulation/pentesters, a platform engineer, and the application security team.

You will focus on the R&D and scaffold design side of automated AI-enabled adversarial testing:

• Design and build scaffolds to automate attacker/threat modeling, attack discovery and exploitation techniques at scale

  • Identify promising attack surfaces and scenarios across Wealthsimple’s stack.

  • Architect and tune agents, prompts, and tool chains that implement real attacker TTPs.

  • Define success metrics and evaluation criteria for automations/ai so we can select and fine tune tooling and model use.

  • Design and iterate on multi-step agent strategies that combine observation, planning, action, and self-learning.

  • Improve effectiveness and automation coverage and reduce unproductive actions and loops.

  • Propose and validate new tools or environment features that enable richer or more realistic attacks.

• Research and design new AI-driven attack strategies and scenarios in anticipation of what adversaries might misuse LLMs to do in future, then help design detections and defensive measures.

• Analyze AI behavior and results to discover systemic weaknesses and strengths and improve platform design / outputs and compensate for weaknesses.

  • Compare different models, prompts, and tool sets on the same scenarios.

  • Measure meaningful outcomes (bugs found, depth of compromise, time-to-finding, false-positive behaviour).

  • Benchmark AI-driven testing against our other tooling and manual test results to understand return on investment and where to invest effort and expertise to best advantage.

• Translate agent outputs into high-quality findings and systemic improvements.

  • Identify high-confidence vulnerabilities and attack paths.

  • Analyze findings to uncover recurring vulnerability types and control gaps, then help us fix them.

  • Understand how agents discovered issues and what that implies for our defences.

  • Share learnings and help build guardrails, detections, systemic framework fixes, libraries, or new agents/experiments.

You’ll help shape the team’s direction through research, proposals, benchmarking, and improved design/implementation.

• Courageously Ambitious – they enthusiastically tackle big audacious goals.

• Deeply Human – they take responsibility for bringing the best out of themselves and others.

• Problem Solvers for scale – they have the ability and resilience to tackle complex issues, find common patterns, design solutions for scale, and see them through.

• Enthusiastic Communicators – they capture and share learnings by default, and are always looking to implement suggestions for improvements and guardrails.

• Embraces change and experimentation – treat campaigns and framework changes as experiments. Thoughtfully define hypotheses, evaluation criteria, and success metrics, then analyze outcomes and share results with the team to guide next iterations.

• 5+ years of experience in offensive security and/or vulnerability research.

• Prior work blending automation with offensive security (e.g., custom tooling, fuzzer integrations).

• Strong technical skills in reading and reasoning about code, infrastructure, and designs.

• Experience building, evaluating, or using LLM- or agent-based systems in any domain.

• A strong curiosity about and openness to AI-augmented workflows:

  • Comfortable iterating on prompts, tools, and agent behaviours.

  • Pragmatic about what AI can and cannot do today.

• Working experience with large language…