85
Techniques Assessed
6
Directly Protected
21
Partially Protected
48
Out of Scope
10
Not Applicable

Methodology

This mapping covers all techniques from the MITRE ATT&CK Enterprise Cloud matrix (IaaS, SaaS, Identity Provider, Office Suite) and the Containers matrix. Techniques appearing in both are listed once. Assessment is based on Phantom's core mechanisms:

  • Mutating admission webhook — intercepts all pod creation, can enforce image policies, block privileged containers, inject sidecars
  • In-memory secret delivery — secrets fetched from EU-hosted OpenBao directly into process memory; never written to etcd, env vars, or volumes
  • mTLS sidecar communication — encrypted channel between sidecar and OpenBao; separate PKI from cloud provider
  • Confidential Computing — when combined with AMD SEV-SNP / Intel TDX nodes, VM memory is hardware-encrypted and inaccessible to the hypervisor

Protection Levels

Protected Phantom directly mitigates this technique by design
Partial Some protection via webhook enforcement, mTLS, or confidential VMs — not the primary purpose
Out of Scope Not addressed — requires other security controls
N/A Technique is irrelevant to Kubernetes secret protection

Phantom is a precision tool, not a broad security platform

Phantom's coverage is narrow but deep. It provides complete protection in the Credential Access and Collection tactics — exactly where secrets are at risk. The 48 "out of scope" techniques are things like account discovery, denial of service, and email spoofing — threats that require entirely different security controls and aren't related to data sovereignty.

Tactics & Techniques

Initial Access — 2 partial, 2 out of scope, 3 N/A

How adversaries gain their first foothold. Phantom doesn't prevent initial access but limits what an attacker can reach once inside.

IDTechniqueMatrixStatusNotes
T1078Valid AccountsCloud ContainerPartialCloud creds don't grant access to OpenBao secrets. Separate auth chain limits blast radius.
T1195Supply Chain CompromiseCloudPartialWebhook can enforce image signing verification and block untrusted images.
T1190Exploit Public-Facing ApplicationCloud ContainerOut of ScopeApplication-level vulnerability. Requires WAF, patching, etc.
T1199Trusted RelationshipCloudOut of ScopeThird-party trust exploitation. Requires vendor management controls.
T1133External Remote ServicesContainerOut of ScopeExposed services like K8s API, dashboards. Requires network policies.
T1189Drive-by CompromiseCloudN/ABrowser-level attack. Not relevant to K8s workloads.
T1566PhishingCloudN/ASocial engineering. Even if creds stolen, they don't unlock OpenBao.
Execution — 4 partial, 3 out of scope, 1 N/A

How adversaries run malicious code. The webhook's admission control provides a strong gating mechanism for container execution.

IDTechniqueMatrixStatusNotes
T1609Container Administration CommandContainerPartialWebhook can restrict kubectl exec. Secrets are in-memory, not in env vars or files.
T1610Deploy ContainerContainerPartialWebhook validates all pod deployments. Can block unauthorized images.
T1204User Execution (Malicious Image)Cloud ContainerPartialWebhook can enforce trusted image registries and signing.
T1677Poisoned Pipeline ExecutionCloudPartialSecrets aren't in CI/CD pipelines. OpenBao delivers only at runtime to authorized pods.
T1053Scheduled Task/JobContainerPartialWebhook intercepts CronJob/Job pod creation. Same admission policies apply.
T1651Cloud Administration CommandCloudOut of ScopeCloud API abuse (gcloud, aws cli). Requires IAM controls.
T1059Command and Scripting InterpreterCloudOut of ScopeCloud API scripting. Requires IAM and audit controls.
T1072Software Deployment ToolsCloudOut of ScopeAbuse of deployment tools like Ansible, Terraform. Requires pipeline security.
T1648Serverless ExecutionCloudN/ALambda/Cloud Functions. Phantom is Kubernetes-only.
Persistence — 2 partial, 5 out of scope, 1 N/A

How adversaries maintain access. Phantom's webhook provides ongoing admission control, and its separate auth chain limits persistence value.

IDTechniqueMatrixStatusNotes
T1525Implant Internal ImageCloud ContainerPartialWebhook can enforce image hash pinning and block tampered images.
T1556Modify Authentication ProcessCloudPartialPhantom uses its own mTLS auth chain to OpenBao. Modifying cloud auth doesn't affect secret access.
T1098Account ManipulationCloud ContainerOut of ScopeCloud IAM manipulation. Requires IAM monitoring.
T1136Create AccountCloud ContainerOut of ScopeNew account creation in cloud or K8s. Requires audit controls.
T1546Event Triggered ExecutionCloudOut of ScopeCloud event triggers (Lambda, Cloud Functions). Requires event pipeline controls.
T1543Create or Modify System ProcessContainerOut of ScopeHost-level process manipulation. Requires node hardening.
T1671Cloud Application IntegrationCloudOut of ScopeOAuth app registration abuse. Requires app governance.
T1137Office Application StartupCloudN/AOffice macro persistence. Not relevant to K8s.
Privilege Escalation — 1 partial, 3 out of scope

How adversaries gain higher-level permissions. Phantom's primary defense here is that even with elevated cloud privileges, secrets remain in a separate EU-controlled domain.

IDTechniqueMatrixStatusNotes
T1611Escape to HostContainerPartialContainer escape gives host access, but with Confidential VMs (SEV-SNP), process memory is hardware-encrypted. Secrets remain protected even after escape.
T1548Abuse Elevation Control MechanismCloudOut of ScopeCloud IAM privilege escalation. Requires least-privilege IAM.
T1484Domain or Tenant Policy ModificationCloudOut of ScopeOrg policy changes. Requires policy monitoring.
T1068Exploitation for Privilege EscalationContainerOut of ScopeKernel/runtime exploits. Requires patching and seccomp profiles.

Confidential Computing upgrade

T1611 (Escape to Host) is the most critical container privilege escalation. With Confidential VMs, even a full host compromise cannot read encrypted process memory — making this a strong protection when combined with Phantom.

Defense Evasion — 5 partial, 6 out of scope, 1 N/A

How adversaries avoid detection. Phantom's webhook and mTLS create independent defense layers that are harder to evade than cloud-native controls.

IDTechniqueMatrixStatusNotes
T1578Modify Cloud Compute InfrastructureCloudPartialAttacker snapshots VM or modifies instances. With Confidential VMs, snapshots contain encrypted memory. Without them, this is a real threat.
T1562Impair DefensesCloud ContainerPartialWebhook is an admission-level defense. Disabling it requires K8s API access, which is auditable. Can be hardened with failClosed policy.
T1656ImpersonationCloudPartialmTLS between sidecar and OpenBao prevents service impersonation for secret retrieval. Certs are EU-managed.
T1550Use Alternate Authentication MaterialCloud ContainerPartialPhantom tokens are short-lived and separate from cloud tokens. Stolen cloud tokens don't grant OpenBao access.
T1612Build Image on HostContainerPartialWebhook can block pods running locally-built images that aren't in trusted registries.
T1211Exploitation for Defense EvasionCloudOut of ScopeZero-day exploitation. Requires defense-in-depth and patching.
T1564Hide ArtifactsCloudOut of ScopeHidden cloud resources. Requires cloud inventory monitoring.
T1070Indicator RemovalCloud ContainerOut of ScopeLog deletion. Requires immutable logging (e.g., external SIEM).
T1666Modify Cloud Resource HierarchyCloudOut of ScopeOrg/folder manipulation. Requires organizational controls.
T1535Unused/Unsupported Cloud RegionsCloudOut of ScopeDeploying to obscure regions. Requires org policy constraints.
T1036MasqueradingContainerOut of ScopeDisguised container names/images. Requires image verification.
T1672Email SpoofingCloudN/AEmail-based attack. Not relevant to K8s infrastructure.
Credential Access — 4 protected, 4 partial, 1 out of scope, 2 N/A ★ Core Protection

This is Phantom's primary battleground

Credential Access is where Phantom delivers its strongest value. 4 techniques are directly protected and 4 more are partially covered. The core design — secrets in process memory only, EU-managed PKI, separate auth chain — addresses the fundamental ways attackers steal credentials.

IDTechniqueMatrixStatusNotes
T1552Unsecured CredentialsCloud ContainerProtectedCORE. Secrets never written to etcd, env vars, configmaps, volumes, or any cloud-accessible store. They exist only in process memory.
T1528Steal Application Access TokenCloud ContainerProtectedOpenBao tokens are short-lived, in-memory, and scoped to specific pod identities. No persistent tokens to steal.
T1555Credentials from Password StoresCloudProtectedSecrets are NOT in any cloud-native secret store (GCP Secret Manager, AWS Secrets Manager, etc.). They're in EU-hosted OpenBao.
T1040Network SniffingCloudProtectedmTLS between sidecar and OpenBao. All secret transit is encrypted with EU-managed certificates.
T1110Brute ForceCloud ContainerPartialOpenBao supports rate limiting and lockout. Separate auth system from cloud IAM.
T1606Forge Web CredentialsCloudPartialPhantom uses a separate credential chain (mTLS certs, not SAML/OAuth). Forging cloud tokens doesn't grant secret access.
T1649Steal or Forge Authentication CertificatesCloudPartialPKI is EU-managed, separate from cloud provider's CA. Attacking cloud PKI doesn't compromise the Phantom cert chain.
T1556Modify Authentication ProcessCloudPartialModifying cloud IdP doesn't affect OpenBao auth. Phantom's auth chain is independent.
T1212Exploitation for Credential AccessCloudOut of ScopeZero-day exploitation of credential systems. Requires patching and hardening.
T1621Multi-Factor Authentication Request GenerationCloudN/AMFA fatigue attacks. Not relevant to machine-to-machine auth.
T1539Steal Web Session CookieCloudN/ABrowser session theft. Phantom doesn't use web sessions for secret delivery.
Discovery — 1 partial, 15 out of scope

How adversaries explore and map the environment. Discovery is largely outside Phantom's scope — these are reconnaissance activities that require network segmentation, RBAC, and monitoring.

IDTechniqueMatrixStatusNotes
T1619Cloud Storage Object DiscoveryCloudPartialSecrets aren't in cloud storage objects. Attackers can enumerate buckets but won't find secrets there.
T1087Account DiscoveryCloudOut of ScopeEnumerating accounts. Requires RBAC restrictions.
T1580Cloud Infrastructure DiscoveryCloudOut of ScopeMapping cloud resources. Requires least-privilege IAM.
T1538Cloud Service DashboardCloudOut of ScopeConsole access. Requires MFA and session controls.
T1526Cloud Service DiscoveryCloudOut of ScopeEnumerating cloud services. Requires IAM restrictions.
T1680Local Storage DiscoveryCloudOut of ScopeBrowsing local storage. Secrets aren't on disk.
T1654Log EnumerationCloudOut of ScopeReading logs. Requires log access controls.
T1046Network Service DiscoveryCloud ContainerOut of ScopePort scanning. Requires network policies.
T1201Password Policy DiscoveryCloudOut of ScopeEnumerating password policies. Low impact for machine auth.
T1069Permission Groups DiscoveryCloud ContainerOut of ScopeMapping RBAC groups. Requires RBAC restrictions.
T1518Software DiscoveryCloudOut of ScopeIdentifying installed software. Requires runtime security.
T1082System Information DiscoveryCloudOut of ScopeOS/hardware enumeration. Low impact for secret protection.
T1614System Location DiscoveryCloudOut of ScopeDetermining geo-location. Not directly relevant.
T1049System Network Connections DiscoveryCloudOut of ScopeNetwork mapping. Requires network policies.
T1613Container and Resource DiscoveryContainerOut of ScopeEnumerating pods/services. Requires K8s RBAC.
Lateral Movement — 3 out of scope, 1 N/A

How adversaries move through the environment. Phantom's isolation model means compromising one component doesn't grant access to secrets in another — but preventing lateral movement itself requires network controls.

IDTechniqueMatrixStatusNotes
T1021Remote ServicesCloudOut of ScopeSSH, RDP, cloud serial console. Requires access controls and bastion hosts.
T1080Taint Shared ContentCloudOut of ScopePoisoning shared storage. Requires integrity monitoring.
T1072Software Deployment ToolsCloudOut of ScopeAbusing deployment tools for lateral movement. Requires pipeline security.
T1534Internal SpearphishingCloudN/AInternal phishing. Not relevant to machine-to-machine infrastructure.
Collection — 2 protected, 2 out of scope, 1 N/A ★ Strong Coverage

How adversaries gather data of interest. Phantom ensures secrets are absent from the locations where attackers typically look — cloud storage, information repositories, and configuration stores.

IDTechniqueMatrixStatusNotes
T1530Data from Cloud StorageCloudProtectedSecrets are never stored in cloud storage (S3, GCS, Azure Blob). Nothing to collect.
T1213Data from Information RepositoriesCloudProtectedSecrets never written to wikis, repos, SharePoint, or other information stores.
T1119Automated CollectionCloudOut of ScopeAutomated scripts collecting data. Application-level concern.
T1074Data StagedCloudOut of ScopeStaging exfiltration data. Requires DLP controls.
T1114Email CollectionCloudN/AEmail harvesting. Not relevant to K8s infrastructure.
Exfiltration — 1 partial, 2 out of scope

How adversaries steal data. Phantom ensures secrets can't be exfiltrated from cloud infrastructure because they aren't stored there — but application-level data exfiltration requires separate controls.

IDTechniqueMatrixStatusNotes
T1537Transfer Data to Cloud AccountCloudPartialAttacker copies cloud data to their account. Secrets aren't in cloud-accessible storage to copy.
T1048Exfiltration Over Alternative ProtocolCloudOut of ScopeData exfiltration via DNS, ICMP, etc. Requires network monitoring.
T1567Exfiltration Over Web ServiceCloudOut of ScopeData exfiltration to external services. Requires egress controls.
Impact — 1 partial, 9 out of scope, 1 N/A

How adversaries disrupt, destroy, or manipulate. Phantom doesn't prevent destructive attacks, but its architecture means secrets survive infrastructure destruction because they're managed externally.

IDTechniqueMatrixStatusNotes
T1486Data Encrypted for ImpactCloudPartialRansomware can't encrypt secrets because they're in EU-hosted OpenBao, not in cloud infrastructure. Secrets survive cloud-side ransomware.
T1531Account Access RemovalCloudOut of ScopeLocking out admins. Requires break-glass procedures.
T1485Data DestructionCloud ContainerOut of ScopeDeleting data. Secrets are safe in OpenBao, but app data requires backups.
T1491DefacementCloudOut of ScopeWebsite/service defacement. Application-level concern.
T1499Endpoint Denial of ServiceCloud ContainerOut of ScopeService disruption. Requires rate limiting and scaling.
T1657Financial TheftCloudOut of ScopeCrypto mining, billing abuse. Requires billing alerts.
T1490Inhibit System RecoveryCloud ContainerOut of ScopeDeleting backups/snapshots. Requires backup protection.
T1498Network Denial of ServiceCloud ContainerOut of ScopeDDoS. Requires cloud DDoS protection.
T1496Resource HijackingCloud ContainerOut of ScopeCryptojacking. Requires resource monitoring.
T1489Service StopCloudOut of ScopeStopping services. Requires HA and access controls.
T1667Email BombingCloudN/AEmail-based disruption. Not relevant to K8s.

Coverage by Tactic

TacticTotalProtectedPartialOut of ScopeN/A
Credential Access114412
Collection52021
Defense Evasion120561
Execution90531
Initial Access70223
Persistence80251
Exfiltration30120
Privilege Escalation40130
Impact110191
Discovery1601150
Lateral Movement40031
Total856214810

Confidential Computing Multiplier

When Phantom is deployed on Confidential VM nodes (AMD SEV-SNP / Intel TDX), several techniques that are "partial" or "out of scope" gain significant additional protection because the hypervisor cannot read process memory:

IDTechniqueWithout Confidential VMsWith Confidential VMs
T1611Escape to HostPartialProtected
T1578Modify Cloud Compute InfrastructurePartialProtected
T1552Unsecured Credentials (memory dump vector)ProtectedProtected+
T1609Container Administration CommandPartialPartial+

With Confidential VMs: 8 protected, 20 partial

Confidential Computing upgrades 2 techniques from partial to fully protected. The webhook can enforce confidential node scheduling via node affinity and taints, making this a zero-configuration upgrade path.

Key Takeaways

Strongest where it matters most

Phantom achieves 73% coverage (4 protected + 4 partial out of 11) in Credential Access — the tactic most directly related to data sovereignty and secret protection. This is not a generic security tool trying to cover everything; it's a focused solution that excels at its core mission.

Complement, don't replace

48 techniques are out of scope. Customers still need: cloud IAM controls, network policies, runtime security (Falco/Tetragon), SIEM/logging, vulnerability management, and DDoS protection. Phantom handles the data sovereignty layer — everything else remains the customer's responsibility.

CLOUD Act specific value

The techniques most relevant to a CLOUD Act compelled disclosure scenario are T1552 (Unsecured Credentials), T1555 (Credentials from Password Stores), T1530 (Data from Cloud Storage), T1578 (Modify Cloud Compute Infrastructure / VM snapshot), and T1213 (Data from Information Repositories). Phantom provides Protected or Partial coverage for all five. With Confidential VMs, all five become fully Protected.

Data Sources

Based on MITRE ATT&CK v16+ (October 2025 update). Enterprise matrix: 216 techniques total, of which the Cloud matrix contains ~89 and Containers matrix contains ~39 unique techniques. Combined and deduplicated: 85 unique techniques relevant to Phantom's operating environment.

  • MITRE ATT&CK Cloud Matrix — attack.mitre.org/matrices/enterprise/cloud/
  • MITRE ATT&CK Containers Matrix — attack.mitre.org/matrices/enterprise/containers/
  • Assessment reflects Phantom's architecture as described in the Security Analysis and Architecture pages