Senior Offensive Security Engineer

At Quorum Cyber, we're on a mission to help good people win. Founded in Edinburgh in 2016, we're one of the fastest growing cyber security companies in the UK and North America, serving over 400 customers on four continents.

We protect organisations against the rising threat of cyber-attacks, enabling them to thrive in an increasingly unpredictable and inhospitable digital landscape.

As a Microsoft-only security house, a Microsoft Solutions Partner for Security, a member of the Microsoft Intelligent Security Association (MISA), and winner of the Microsoft Security MSSP of the Year 2025 award, we offer a unified security ecosystem comprised of innovative services, all delivered through our customer platform, Clarity.

In September 2024, Quorum Cyber acquired Canada-based, Microsoft Solutions Partner for Security, Difenda. This was closely followed in December 2024 by the acquisition of US-based, Kivu Consulting, a global cyber security firm with world-leading incident response capabilities.

In this role you will apply your leadership, innovative thinking, curiosity, and existing deep technical expertise to help Quorum Cyber close the distance between AI speed and efficiency, and human insight. Penetration testing, API and web application testing, and red teaming are delivered the traditional way: a skilled human, a scope document, a fixed number of days, a report at the end.

The work is excellent, but it does not scale. Clients want continuous assurance; not la snapshot.

This role exists to change that. You will run real engagements and, from inside that work, build the agentic AI that takes them over stage by stage: reconnaissance, enumeration, attack-path discovery, exploitation of known vulnerability classes, evidence capture, triage, first-draft reporting.

• Lead network, web application, API, cloud, and full-scope red team engagements.
• Own them end to end: scoping, rules of engagement, authorisation, execution, evidence, reporting, client debrief.
• Set the quality bar and be the escalation point on a hard target.

• Build agentic systems that perform discrete stages of a test autonomously, starting narrow and expanding as reliability is proven.
• Build the evaluation harness before the agent:
• Engineer for the failure modes that matter here: false positives and negatives, non-determinism, scope escape, destructive actions, runaway cost. Instrument accuracy, coverage, cost, and hours saved.

• Work with service management and Product Management to turn working automation into a repeatable, multi-tenant offering with defined SLAs and pricing.
• Enforce scope and authorisation in code rather than in a document: target validation, blast-radius limits, kill switches, prohibited-action lists, full audit trail.
• Define where the human stays in the loop, and mentor the team in both directions.

Strength in all three areas. Depth in offensive security and demonstrable AI agent-building are both non-negotiable.

• Around 7 years hands-on, including at least 4 delivering client-facing engagements.
• Network testing: external and internal, Active Directory attack paths, privilege escalation, lateral movement, post-exploitation.
• Web and API testing: the OWASP Top 10 and, more importantly, what it misses, meaning business logic flaws, authentication and session weaknesses, and multi-step chains.
• Red teaming: objective-based operations covering initial access, command and control, EDR evasion, and purple-team work, against a client actively trying to catch you.
• Excellent written English, with redacted reports or a willingness to sit a writing exercise. The report is the product.

• The part we care most about, and the part candidates most often overstate. We want evidence of building, not of using.
• You have shipped a non-trivial LLM agent: something that plans, calls real tools, handles errors, and finishes a multi-step task unsupervised. A private repository or a work project you can describe in detail counts.
• You are fluent with tool calling, structured output, the Model Context Protocol, and at least one agent framework or SDK (Lang Graph, CrewAI, the OpenAI or Claude Agent SDKs, PydanticAI, or a hand-rolled loop). We are framework-agnostic and interested in your reasoning.
• You can describe an evaluation…