Multi-Agent AI Engineer — Red Team Reasoning

Cyrion AI; Greater London; Remote; GBP 90,000 - 130,000; Full time

Benefits offered by this job

Early equity

Full health coverage

$2,000/yr learning budget

Hardware of your choice

Job summary

Cyrion AI in Greater London is building a swarm of coordinated LLM-driven agents that reason about attack surfaces the way a human red teamer would. You’ll own pieces of that reasoning loop—from hypothesis generation to confidence scoring to validation handoff.

This is a production system with real customers; you’ll design and train prompts, build evaluation harnesses, and improve agent coordination across the swarm.

Qualifications

• 3+ years building production ML or LLM-backed systems.
• Strong Python; comfortable with modern agent/orchestration frameworks.
• Experience with retrieval, tool-calling, or multi-step planning architectures.
• A security or adversarial-ML background is a strong plus but not required — we will pair you with offensive researchers.

Responsibilities

• Design and train/prompt-engineer agent components for recon synthesis, hypothesis generation, and exploit chaining.
• Build evaluation harnesses that measure agent accuracy against known-ground-truth environments.
• Improve agent-to-agent coordination and shared memory across the swarm.
• Reduce false-positive rate by tightening the confidence-scoring and validation handoff logic.
• Ship incrementally — this is a production system with real customers, not a research sandbox.

Skills

Python

LLM-backed systems

Agent frameworks

Retrieval

Multi-step planning

Security/adversarial ML

Tools

Modern orchestration frameworks

Job description

Cyrion AI in Greater London is building a swarm of coordinated LLM-driven agents that reason about attack surfaces the way a human red teamer would. You’ll own pieces of that reasoning loop—from hypothesis generation to confidence scoring to validation handoff.

This is a production system with real customers; you’ll design and train prompts, build evaluation harnesses, and improve agent coordination across the swarm.