We were founded on the observation that the attack surface of a production AI system is not a list of CVEs — it is the gap between what the model was trained to do and what an adversary can make it do. Closing that gap requires offensive capability: the ability to find exploits before they are found for you. LogicLeak is a boutique practice built entirely around that capability.
How an engagement runs.
Scoping call
We spend 60 minutes understanding your AI architecture, trust boundaries, data flows, and existing security controls. We agree on attack surface scope, success criteria, and rules of engagement.
Threat modelling
We build a structured threat model against your specific system — not a generic template. Every attack path is grounded in your actual deployment: your RAG index, your tool access list, your system prompt, your users.
Active red-team campaign
We execute the attack plan. Automated scans establish baseline coverage; manual campaigns target the high-value paths. We document every finding with a proof-of-concept and reproduction steps.
Findings review
We present findings in a structured debrief: severity ranking, attack chain walkthrough, business impact framing. We separate findings that require architectural changes from those addressable at the prompt level.
Remediation + re-test
We provide a prioritised remediation playbook. Surgery and Retainer engagements include implementation support and a re-test to verify each critical finding has been resolved.
Operating principles.
Transparency
Every finding ships with a reproduction case. Nothing is claimed without evidence. Our reports are written to be acted on, not filed.
Reproducibility
Exploits are documented to the level that your engineering team can re-run them. If we can't reproduce it reliably, we don't report it as a confirmed finding.
Responsible disclosure
When our research surfaces novel attack techniques with broad industry impact, we follow coordinated disclosure — notifying affected vendors and giving reasonable remediation time before publication.
Containment
Red-team activities are conducted in isolated environments where possible. Where production testing is required, we agree blast-radius limits in advance and maintain kill-switch capability throughout.
Continuity
Retainer clients receive continuous adversarial coverage — not a point-in-time report that ages out. As your AI systems evolve, our assessment evolves with them.
Independence
We are not affiliated with any model provider, cloud platform, or AI tooling vendor. Our recommendations are based solely on what reduces your risk.
LogicLeak operates under InFrontWebs Ltd, registered in England & Wales. All engagements are subject to a signed master services agreement.
Work With Us →