home / services
Services
Independent security work on the boundaries this portfolio covers: AI and web security, architecture and threat modelling, authorised adversarial testing, detection engineering and supply-chain integrity.
How I work
scoped · documented · verifiableModel the trust boundary before touching a configuration, prove every finding with the smallest reproduction that fits in an isolated environment, and report on a pre-agreed clock. Every engagement is scoped and authorised in writing first — the scope page says what that means, and the approach on the overview page shows how I work.
AI Security & Adversarial Assurance
agents · prompt injection · RAG isolationSecurity for systems built around models, where agents, retrieval pipelines and tooling trust input that should be treated as untrusted. This is the area covered by the AI security research on the research page.
Agent and prompt-injection testing
Direct and indirect prompt injection across agent and tool boundaries, each path reported with a minimal reproduction and a fix.
RAG and data-flow isolation
What retrieved content is allowed to do once it reaches the model, and where document trust should stop.
AI authorization review
Tool and action boundaries, delegated permissions and unsafe model loading in AI infrastructure.
Security Architecture & Threat Modelling
trust boundaries · controls · validationDesign reviews that start at the trust boundary and end with controls that can be tested. Threat models are published as markdown in the threat model repository.
Threat modelling
A written model of what the system assumes, what can cross the boundary, and what happens when it does.
Secure design review
Architecture, trust boundaries and data flow reviewed before implementation locks the design in.
Control validation
Security controls checked against the model, so the design and the deployed reality stay the same.
Adversarial Security Assessment
authorised · identity · attack pathsAuthorised testing of web and API surfaces, identity infrastructure and the attack paths between them, reported with reproductions and fixes. Sample findings are published after disclosure on the findings page.
Web and API testing
Authentication, access control, SSRF and injection on modern stacks, scoped and agreed before testing starts.
Identity and privilege escalation
IAM misconfigurations, role boundaries and the escalation paths that connect one identity to too many permissions.
Attack-path analysis
The shortest paths from a foothold to the crown jewels, so the cheap fixes come first.
Detection Engineering & SOC
Sigma · Splunk · KQL · MITRE ATT&CKDetection rules that document their own false positives, and SOC work that treats triage as an engineering problem. The rule library is published on the detections page.
Custom detection rules
Sigma, Splunk SPL and Microsoft Sentinel KQL, each mapped to MITRE ATT&CK with its false positives written down.
Alert triage and tuning
Work the queue, cut the noise, and tune rules against what an analyst actually sees.
SOC process review
Escalation paths and runbooks reviewed against real incident timelines.
Software Supply-Chain Security
CI/CD · build isolation · provenanceMaking builds reproducible, isolated and verifiable from pipeline to artifact. A worked example — a leaked pipeline token publishing a tampered build — is on the incident response page.
CI/CD security review
Pipeline tokens, privileged steps and build context, checked the same way as any other attack surface.
Build isolation and reproducibility
Deterministic builds in isolated environments, so what is reviewed is what is deployed.
Artifact integrity and provenance
Signing, attestation and source-to-artifact traceability reviewed end to end.