Yinkozi
Contact
services / cloud security

Cloud security as attack-path analysis — not a CSPM scan repackaged.

Manual cloud-security assessment of AWS, Azure, and GCP estates. IAM-driven attack-path analysis, cross-account misconfiguration, identity-federation abuse, custom infrastructure-as-code review. We are not selling you a Cloud Security Posture Management dashboard.

01 / why CSPM is not enough

Misconfigurations are not the interesting attacks.

Cloud security posture management tools find S3 buckets that are public, security groups that allow 0.0.0.0/0, and IAM roles that have admin privileges. Useful — and largely solvable with policy-as-code.

The interesting attacks against modern cloud estates are not single-misconfiguration findings. They are attack paths: a chain of three or four legitimate-looking permissions, across two or three accounts, that adds up to an undetected privilege escalation. CSPM tools do not see those, because each individual step is allowed.

We model the customer's IAM graph manually, against the customer's real workloads, and find paths an attacker would walk that the customer's tooling does not register as findings.

Cloud security — IAM attack path crossing three accounts — iam attack path · cross-account — attacker traversal ACCOUNT 01 Identity / SSO role Federated user role CI service role · over-privileged Admin-via-OIDC (entry point) Initial foothold via misconfigured GitHub-Actions OIDC trust policy. ACCOUNT 02 Staging role Deploy role Cross-account assume AssumeRoleWithWebIdentity trust policy too permissive, accepts production identifier. ACCOUNT 03 Production role Workload role Read-only audit target reached KMS · customer-data store Workload role allows decrypt on customer-managed keys — no CSPM tool flagged the chain. 3-step path · each step individually allowed · CSPM tools see no finding
02 / what we test

The shape of the work.

IAM attack-path analysis

Manual review of IAM roles, policies, trust relationships, and permission boundaries. Multi-step privilege escalation, cross-account abuse, identity-federation walks.

Cross-account & multi-tenant

Pivot analysis across organizational boundaries, tenant-isolation review, shared-resource abuse, payer-account exposure, control-tower drift.

Identity federation

SSO and SCIM-pipeline abuse, OIDC and SAML protocol-level attacks, AssumeRoleWithWebIdentity misuse, GitHub-Actions OIDC misconfiguration.

Secrets & key management

KMS / Key Vault / Cloud KMS audit, secret-store boundary review, access-pattern analysis, IMDS abuse, container-runtime credential leakage.

Infrastructure-as-code

Terraform, CloudFormation, ARM, Pulumi review. Custom analyzers for the customer's modules and patterns where commercial IaC scanners do not have context.

Container & Kubernetes

EKS / AKS / GKE node-isolation review, RBAC analysis, admission-controller bypass, service-mesh policy review, supply-chain integrity.

Serverless & eventing

Lambda / Functions trust-boundary review, EventBridge and Pub/Sub policy review, S3-event chain abuse, workflow-orchestration misuse.

Data plane

Database access-pattern review, data-lake permission analysis, encryption-at-rest evaluation, customer-managed-key model audit.

Detection-stack review

CloudTrail / Audit Logs / Activity Logs coverage analysis, log-pipeline integrity, detection-rule design review, blue-team validation.

03 / what we don't do

The shape we don't ship.

  • No reselling CSPM

    Wiz, Orca, Prisma Cloud, Lacework — useful tools, owned by the customer, not by us. We do not bill for re-formatting their dashboards.

  • No CIS-benchmark deliverable

    The CIS benchmarks are a useful baseline. Customers run them. We do not bill expert hours to read them back.

  • Local LLMs as tooling support

    Cloud-attack-path analysis depends on graph reasoning over the customer's specific IAM topology. Current models hallucinate paths and miss context — manual remains the right default. We do use local LLMs we run in-house to summarise large policy documents, accelerate IaC analyzer development, and pattern-match across our customer-anonymised corpus. Customer IAM material never goes to a commercial SaaS model.

04 / engagement shape

Scoped to the estate, not to a calendar.

Single-account / single-cloud assessment. One AWS account or Azure subscription, full IAM and resource review, attack-path analysis.

Multi-account / multi-cloud. AWS Organizations, Azure tenancy hierarchies, GCP folder structures. Cross-account pivot analysis.

IaC-pipeline review. Standalone or paired. Terraform / CloudFormation modules, custom-policy analyzers, drift detection, deployment-pipeline integrity.

Continuing engagement. Quarterly review as the cloud estate evolves — new service, new region, new account, new attack surface.

Duration depends on the estate's size and the depth of coverage. We scope against the actual footprint.

05 / start a conversation

Cloud is where most modern compromises happen.

email