AI Security & Cloud Governance
Most organisations running AI workloads in AWS have reviewed infrastructure security but never specifically assessed AI risk — prompt injection, inference endpoint exposure, IAM for Bedrock and SageMaker. Separately, cloud misconfiguration and compliance obligations remain a persistent gap. We address both: a structured AI security review, and a governance assessment mapped to the frameworks your context requires.
Sound familiar?
"We’re using Bedrock in production. The security team signed off on the infrastructure, but nobody has looked specifically at the AI risk surface — what the model roles can actually do, whether inference endpoints are exposed, how we’d detect a prompt injection attempt."
"Security Hub has been running for months. There are over 600 findings in it. Nobody knows which are critical for Essential Eight, which are false positives, and which we’d actually fail an assessment on."
"We’re in financial services and APRA CPS 234 requires us to demonstrate information security capability across our cloud environment. We’ve never formally mapped our AWS posture to those requirements."
What we do
AI-specific risk and cloud governance sit in different parts of your risk register but share the same environment. We run them as separate, scoped engagements — or together where it makes sense.
AI workloads carry a risk surface that standard audits miss — prompt injection, data leakage through inference endpoints, overpermissioned service roles, and third-party model integration risk.
Surface what’s misconfigured, unmonitored, or drifted — then map every gap to the frameworks you need to evidence against.
The engagement
Whether we’re reviewing AI workloads, cloud governance, or both — the engagement follows the same structure. Scope first, then tooling-first review, then a report your team can act on immediately.
Phase 01
Phase 02
Phase 03
What we assess against
Most engagements are driven by a specific obligation — a compliance deadline, an insurer requirement, or a government contract. We agree the right framework in Scope, and where two overlap we map findings to both so you get full coverage from a single engagement.
AI security
AI workloads carry a risk surface that standard cloud reviews don’t cover. We assess model roles, inference endpoints, data access paths, and integration architecture against known AI attack patterns.
Bedrock · SageMaker · IAMStructured testing of AI inputs for injection surfaces, jailbreak vectors, and unintended data disclosure. Relevant for any customer-facing or third-party integrated AI feature.
AI riskCloud governance & compliance
Cyber insurance renewals, government contract requirements, ASD-regulated environments. The 2024 model requires timestamped, system-generated evidence — we produce it.
ML1 · ML2 · ML3Information security requirements for APRA-regulated financial services entities. We map your AWS environment against CPS 234 controls and produce the evidence your prudential audit requires.
Financial servicesAWS-native baseline, built into Security Hub. Covers the controls AWS considers minimum for a production workload — good starting point and widely accepted by enterprise customers and auditors.
AWS FSBPStructured review across identity, detection, infrastructure protection, data protection, and incident response. Can be run standalone or as part of a full six-pillar WAR.
AWS WAFRemediation
The report is a complete, standalone deliverable. Some customers take it and remediate internally. Others want us to execute the fixes. Either way, what gets remediated is entirely your choice — there is no pressure to take more than you need.
Step 01
After the report walkthrough, you review the remediation task list. Each item has a severity, a plain-language description of the fix, and an effort estimate. Nothing is ambiguous.
Step 02
You pick the items — by severity tier, by framework requirement, or individually. We scope a fixed price for each selected item before any work starts. No open-ended remediation retainers.
Step 03
We implement the fixes, run the tooling again to confirm closure, and produce an updated evidence package. The delta report shows before-and-after posture — the document your auditor or insurer needs.
Why Nuvrix
Standard cloud security reviews don’t cover AI risk. We test specifically for prompt injection surfaces, overpermissioned model roles, inference endpoint exposure, and data leakage paths — the attack surface that comes with running AI in AWS.
Findings are mapped to the framework control they breach. Not a general recommendation. A specific gap against a specific requirement — which is what auditors, insurers, and regulators ask for.
The 2024 Essential Eight assessment model requires timestamped evidence that’s system-generated, not manually compiled. Our tooling produces it automatically. Screenshots and spreadsheets no longer meet the bar — our evidence does.
Automated tooling covers the environment in full — every account, every region. Manual review fills the gaps tooling cannot reach. The result is a complete picture, not a sample-based assessment that misses what it doesn’t look for.
The report stands on its own. If you remediate internally, we’ve given you everything you need. If you want us to fix items, we scope each one at a fixed price before starting. No retainer, no scope creep.
We’re reviewing the environment we know how to build. We understand the context behind findings — when something is a genuine gap versus a deliberate architectural decision — and our remediation is AWS-native, not generic.
Tell us what’s driving the review — AI workloads in production, an upcoming audit, a compliance obligation — and we’ll scope a Define engagement. You’ll know what the review covers and what it costs before committing to anything.