BlastMap
An identity-centric threat mapping framework for cloud AI infrastructure, mapping MITRE ATLAS attack tactics to machine identity over-privilege in AWS environments.
BlastMap addresses a growing blind spot in cloud AI security: machine identities IAM roles, service accounts, and instance profiles are consistently over-privileged in AWS environments hosting AI workloads, creating a wide blast radius when compromised.
The framework systematically maps MITRE ATLAS adversarial ML tactics to specific IAM misconfiguration patterns, enabling security teams to prioritize identity remediation based on realistic attack paths rather than theoretical vulnerability scores.
"Most cloud AI security frameworks focus on model protection. BlastMap focuses on what attackers actually target first: the identity layer that gives them access to everything else."
-
01
Attack Surface EnumerationCatalogued MITRE ATLAS tactics relevant to cloud-hosted AI pipelines, training data poisoning, model inversion, supply chain attacks — and mapped each to corresponding AWS IAM primitives (roles, policies, SCPs).
-
02
Over-Privilege Pattern AnalysisAnalyzed common IAM over-privilege patterns in ML environments: wildcard S3 permissions on SageMaker roles, cross-account trust misconfigurations, and excessive Lambda execution permissions.
-
03
Blast Radius ScoringDeveloped a scoring methodology combining CVSS-like severity with lateral movement potential, quantifying how far an attacker can move from a single compromised machine identity.
-
04
Regulatory Compliance MappingCross-referenced remediation recommendations with NIST 800-53 control families, SOC 2 Trust Services Criteria, and EU AI Act requirements for high-risk AI systems.