Job title: Principal Engineer Associate - int, Agentic AI and Security
Location: London, England, United Kingdom · Hybrid · Contract · Mid-Senior level
Birmingham, UK. We are open to global applications.
About NeoFelis Labs
NeoFelis Labs @NeoFelisLabs is a UK deep tech company building assurance infrastructure for AI agents.
Organisations across regulated industries want to deploy agents. Their security functions cannot authorise it, because nobody can answer a basic question with evidence: is this specific agent safe to run in this specific context, and how would we know if that changed?
We are building the layer that answers it. Our methodology comes from functional safety and assurance case practice in automotive and aerospace, where proving a system safe has been a formal discipline for three decades. We are bringing that rigour to agentic AI, where it does not yet exist.
The role
This is our early lifecycle technical hire. You will own the architecture and build the first working system, reporting directly to the CxO and working alongside the technical co-founders.
The six workstreams
Prioritise and design the architecture that holds together.
- Assurance case generation. Automated production of a defensible, continuously updated assurance case for a specific agent in a specific deployment context. Goal Structuring Notation applied to agentic systems.
- B- radius analysis. Mapping what an agent could actually reach given its credentials, tool access and network position, and quantifying worst case impact.
- Behavioural drift detection. Continuous monitoring against a declared behavioural envelope. Detecting scope escalation, unexpected tool invocation, capability creep.
- Agent supply chain assurance. Provenance, integrity and vulnerability tracking across an organisation's agent tooling and connector estate.
- Incident forensics and replay. When an agent causes harm, reconstructing what it saw, what it decided, and why.
- Authorisation console. The interface a security officer uses to make a documented go or no go decision, with a full audit trail.
Academic background
We expect a degree in computer science, software engineering, electronic engineering, mathematics, physics or a closely related discipline. A postgraduate qualification in security, machine learning, formal methods or safety critical systems is welcome but not required. A strong self taught engineer with the right production record will be taken as seriously as a doctorate.
Specialisms, and what evidence looks like
You will/May not have all of these. We are looking for genuine depth in two or three, and working familiarity with the rest.
Agent architecture and orchestration. You have built systems where an LLM plans, calls tools and acts with consequence. You have designed permission models, sandboxing or human in the loop checkpoints. You have opinions about what breaks in production, formed by watching it break. Neuromorphics/Spiking Tech is preferred but may bot be necessary.
Security engineering. Threat modelling, privilege boundaries, identity and access architecture, secrets management. You have secured systems that mattered, in cloud or on premise, and you have been through a real incident.
Adversarial machine learning. Prompt injection, jailbreaking, tool abuse, data exfiltration through model outputs. You have attacked models, or defended them, and can describe what actually worked.
Formal methods and verification. Model checking, temporal logic, runtime verification, static analysis. Anything from TLA+ or Alloy to invariant checking in production. You can tell the difference between a property you can prove and one you can only monitor.
Evaluation and benchmarking. You have built harnesses that produce reproducible, defensible measurements rather than favourable ones. You understand why most published evaluations do not replicate.
Safety critical and regulated systems. Experience with ISO 26262, DO-178C, IEC 61508, IEC 62304 or equivalent. You have written or defended a safety argument to an assessor. This is unusual in AI and disproportionately valuable to us.
Observability and instrumentation at scale. Distributed tracing, structured telemetry, anomaly detection over behavioural data. You have instrumented something complex without drowning it in noise.
Open source. You have maintained a project, shipped to a community, or led a specification. Our core will be open source and that is a strategic decision, not a marketing one.
Requirements
- Substantial experience building and operating production software
- Deep working knowledge of agent architectures, tool use and permission models
- Strong security engineering foundation
- Able to work at an early stage with real ambiguity and limited structure
(Eligible for UK security clearance)
- Experience in UK Gov and Social services /NHS processes, working knowledge of bids/ competetions applications
Terms
Three/Two days per week, contract initially, with the intention of moving/evolving optionally to a permanent type of contract role on funding. Hybrid, London based.
Skills: AI Safety · Cybersecurity · Machine Learning · Python · Distributed Systems · Software Architecture · Security Engineering · Large Language Models · Formal Verification