AI security research lab

Your agents will be attacked.

Adversarial testing, secure-by-design engineering, and autonomous security products for agent-native security teams and products.

Our thesis

Security as a living system.

Autopoiesis — from the Greek autos (self) and poiesis (creation) — is how biologists Maturana and Varela defined life itself: a system that continuously produces and repairs the components that keep it alive, inside a boundary it makes for itself. We believe cybersecurity has to work the same way.

01 Structural coupling

Evolve

A living system endures by continuously adapting to its environment without losing what makes it itself. Security for autonomous systems has to do the same — sensing, learning from, and adapting to a threat landscape that shifts faster than any release cycle, rather than standing behind a fixed perimeter.

02 Operational closure

Contain

A cell produces a membrane that separates self from non-self and keeps its internal processes coherent. Agentic systems need the same: explicit trust boundaries, least privilege, and sandboxing — so a single compromise stays contained and the blast radius is bounded by design.

03 Self-production

Sustain

Life persists by constantly regenerating its own components. Security is not a one-off audit but a self-sustaining loop — test, learn, harden, and regenerate defences continuously, as the system and its adversaries co-evolve.

Our service offerings

Securing autonomous systems.

We build security products and work directly with the businesses shipping agentic systems — from hands-on red teaming and secure engineering to deployable capability your own cybersecurity team can run at scale.

01

Red teaming agentic systems

Adversarial testing of autonomous agents — before an attacker gets there first.

02

Building secure agentic systems

Agentic architectures with security designed in, not bolted on.

03

Advisory & architecture review

Architecture review, threat modelling, and ongoing security advisory.

04
In development

Reinforcement learning as a service (RLaaS)

End-to-end RL — environment design, reward modelling, training, and evaluation.

05
In development

Private Cyber-Reasoning-Systems (PCRS)

A private autonomous cyber capability that autonomously detects, patches and verifies resolution to your internal source code vulnerabilities.

Always researching…

More to come

New attacks and defences for agentic systems ship as the frontier moves.

Get in touch!

Send us an email, or reach out out directly on LinkedIn

info@autopoiesis.uk