1. Resources
  2. Detection rules
  3. Key Vault Secrets, Keys or Certificates Read in Bulk

Key Vault Secrets, Keys or Certificates Read in Bulk detection rules

An identity with data access to an Azure key vault reads many of its secrets in a short time, or exports its objects as backup files. The reads appear only in the vault's own resource log, not in the Azure Activity log.

Platform
Azure
Severity
High
Log source
Azure Key Vault resource logs (AuditEvent)
Last updated
Rule formats
Sigmaand queries for: SplunkMicrosoft Sentinel
What an attacker needs
Data access to the vault's secrets, or control-plane rights such as Owner or Key Vault Contributor that let the attacker grant that access to an identity they control.
Impact
The credentials stored in the vault, which open databases, storage and privileged identities and keep working until each one is rotated at its source.
Response
Remove the caller's access to the vault, rotate every secret that was read at its source and reissue certificates whose private key was exposed.
Mitigation
Export the AuditEvent log of every vault, grant data roles per application identity and close public network access to the vault.
Key Vault operations
SecretGetSecretBackupKeyBackupCertificateBackup

How the attack works

An Azure key vault holds the credentials that applications and pipelines use: connection strings, API keys, service principal secrets, and certificates with their private keys. Reading a secret is a data-plane request to the vault (SecretGet), and it is the same request whether an application fetches the one secret it needs or an attacker reads every secret the vault holds. Microsoft documents that a Key Vault certificate has an addressable secret with the same name, and that an exportable certificate is retrieved from that secret together with its private key, so reading secrets also yields certificates. Keys are different: the vault never returns private or symmetric key material, but a key, a secret or a certificate can be downloaded as a backup blob (KeyBackup, SecretBackup, CertificateBackup) that can be restored into another key vault in the same subscription and Azure geography.

An attacker who holds data access to a vault, or who has just granted it to an identity under their control, lists the vault and reads its secrets one after another. Microsoft's report on the threat actor it tracks as Storm-2949 (May 2026) describes this: within four minutes the actor changed the access configuration of a key vault and read dozens of its secrets, among them database connection strings and identity credentials. None of this is in the Azure Activity log, which records control-plane operations. The reads are in the vault's resource log (category AuditEvent), which exists only when a diagnostic setting on the vault exports it. Each record carries the operation name, the result, the caller's IP address, the vault and the identifier of the object that was returned. The rules on this page count different secrets per caller address and vault, because the identity claims of the record are not mapped as fields.

What an attacker needs

  • Data-plane access to the vault: an access policy with the get permission on secrets, or an Azure role with Microsoft.KeyVault/vaults/secrets/getSecret/action such as Key Vault Secrets User, Key Vault Secrets Officer or Key Vault Administrator. For a backup, the backup permission on the object type.
  • Or control-plane rights that lead there. On a vault that uses access policies, Microsoft documents that an identity with Microsoft.KeyVault/vaults/write can add an access policy for itself. On a vault that uses Azure RBAC, Owner and User Access Administrator can assign a data role.
  • Network access to the vault. A vault that allows public network access can be read from anywhere with a valid token.

Impact

Every credential the vault holds: database connection strings, storage and API keys, secrets of service principals and certificates with exportable private keys. Each of them opens another system, often with more privilege than the identity that read the vault. The credentials keep working after the attacker loses access to the vault, until each one is rotated at its source. A backup blob that was restored into another vault is an independent copy that the original vault cannot revoke.

Detection rules

Many Different Secrets Read From One Key Vault by One Address

Counts, per caller IP address and key vault, how many different secrets were read successfully (operation SecretGet, counted by the object identifier the record returns in Id). An application reads the few secrets it needs, often the same one many times. Ten or more different secrets from one address within 10 minutes mean that one client went through the content of the vault, which is what an attacker with data access does to collect credentials. The identifier includes the secret version, so several versions of one secret count separately. The identity claims of the record are not mapped as fields, so the rule groups by address and vault, and the caller has to be named from the matching records. The count and the window are starting points. The log exists only where a diagnostic setting on the vault exports the AuditEvent category.

Fires when one combination of CallerIpAddress and ResourceId has at least 10 different Id values within 10 minutes.

Threshold ruleLevelMediumStatusExperimentalLog sourceAzure Key Vault resource logs (AuditEvent)
title: Key Vault Secret Read (Threshold Base Event)
id: efcdcc9c-7abf-459f-b530-37840ae033ad
name: key_vault_secret_read
status: experimental
description: >-
  Base event of the threshold rule in this file. A successful read of a secret
  value from an Azure key vault (operation SecretGet in the vault's AuditEvent
  resource log). On its own this is what every application that uses the vault
  does.
logsource:
  product: azure
  service: keyvault
detection:
  selection:
    OperationName: SecretGet
    ResultType: Success
  condition: selection
level: low
---
title: Many Different Secrets Read From One Key Vault by One Address
id: df7cc0f7-327b-4f58-bba4-3be6934490f9
status: experimental
description: >-
  Counts, per caller IP address and key vault, how many different secrets were
  read successfully (operation SecretGet, counted by the object identifier the
  record returns in Id). An application reads the few secrets it needs, often
  the same one many times. Ten or more different secrets from one address
  within 10 minutes mean that one client went through the content of the vault,
  which is what an attacker with data access does to collect credentials. The
  identifier includes the secret version, so several versions of one secret
  count separately. The identity claims of the record are not mapped as
  fields, so the rule groups by address and vault, and the caller has to be
  named from the matching records. The count and the window are starting
  points. The log exists only where a diagnostic setting on the vault exports
  the AuditEvent category.
references:
  - https://attack.mitre.org/techniques/T1555/006/
  - https://learn.microsoft.com/en-us/azure/key-vault/general/logging
  - https://learn.microsoft.com/en-us/azure/azure-monitor/reference/tables/azkvauditlogs
  - https://learn.microsoft.com/en-us/azure/defender-for-cloud/alerts-azure-key-vault
author: Forestall
date: 2026-11-18
tags:
  - attack.credential-access
  - attack.t1555.006
correlation:
  type: value_count
  rules:
    - key_vault_secret_read
  group-by:
    - CallerIpAddress
    - ResourceId
  timespan: 10m
  condition:
    field: Id
    gte: 10
falsepositives:
  - Applications and configuration providers that load every secret of the vault when they start, at each restart or scale-out. Exclude their addresses, or raise the count above what they read.
  - Backup jobs, secret scanners and migration scripts that go through a vault on purpose. Exclude their addresses once confirmed.
  - The values of 10 different secrets and 10 minutes are starting points, not tested limits. A vault with fewer secrets than the count can never match, so set the count per vault size where you can.
level: medium

False positives

  • Applications and configuration providers that load every secret of the vault when they start, at each restart or scale-out. Exclude their addresses, or raise the count above what they read.
  • Backup jobs, secret scanners and migration scripts that go through a vault on purpose. Exclude their addresses once confirmed.
  • The values of 10 different secrets and 10 minutes are starting points, not tested limits. A vault with fewer secrets than the count can never match, so set the count per vault size where you can.

Many Key Vault Requests Denied for One Address

Counts, per caller IP address and key vault, the requests that the vault answered with the HTTP status Forbidden. Microsoft documents that ResultSignature holds the HTTP status of the request, such as OK or Forbidden. Twenty or more denied requests from one address within 10 minutes mean that a client keeps asking for objects it has no right to, which is what an attacker does who tries a stolen identity against a vault, or who has access to one object type and tries the others. It is also what a misconfigured application does, so the rule is a pointer for review and not proof of an attack. The count and the window are starting points. The log exists only where a diagnostic setting on the vault exports the AuditEvent category.

Fires when one combination of CallerIpAddress and ResourceId has at least 20 matching events within 10 minutes.

Threshold ruleLevelLowStatusExperimentalLog sourceAzure Key Vault resource logs (AuditEvent)
title: Key Vault Request Denied (Threshold Base Event)
id: 6f7f4cfa-9028-4b2b-8704-ddd508e7f2d8
name: key_vault_request_forbidden
status: experimental
description: >-
  Base event of the threshold rule in this file. A request to an Azure key
  vault that was answered with the HTTP status Forbidden, as recorded in the
  ResultSignature field of the vault's AuditEvent resource log. A single
  denied request is a normal result of a missing permission or a firewall
  rule.
logsource:
  product: azure
  service: keyvault
detection:
  selection:
    ResultSignature: Forbidden
  condition: selection
level: low
---
title: Many Key Vault Requests Denied for One Address
id: dd27d1c1-e236-46db-8a98-49e52a3a29a5
status: experimental
description: >-
  Counts, per caller IP address and key vault, the requests that the vault
  answered with the HTTP status Forbidden. Microsoft documents that
  ResultSignature holds the HTTP status of the request, such as OK or
  Forbidden. Twenty or more denied requests from one address within 10 minutes
  mean that a client keeps asking for objects it has no right to, which is
  what an attacker does who tries a stolen identity against a vault, or who
  has access to one object type and tries the others. It is also what a
  misconfigured application does, so the rule is a pointer for review and not
  proof of an attack. The count and the window are starting points. The log
  exists only where a diagnostic setting on the vault exports the AuditEvent
  category.
references:
  - https://attack.mitre.org/techniques/T1555/006/
  - https://learn.microsoft.com/en-us/azure/key-vault/general/logging
  - https://learn.microsoft.com/en-us/azure/key-vault/general/monitor-key-vault
  - https://learn.microsoft.com/en-us/azure/defender-for-cloud/alerts-azure-key-vault
author: Forestall
date: 2026-11-18
tags:
  - attack.credential-access
  - attack.t1555.006
correlation:
  type: event_count
  rules:
    - key_vault_request_forbidden
  group-by:
    - CallerIpAddress
    - ResourceId
  timespan: 10m
  condition:
    gte: 20
falsepositives:
  - An application or a pipeline whose access policy or role assignment was removed, or whose address is no longer allowed by the vault firewall, and which keeps retrying. Fix the access or exclude the address.
  - Inventory and compliance scanners that probe every vault with an identity that has no data access.
  - The values of 20 requests and 10 minutes are starting points, not tested limits.
level: low

False positives

  • An application or a pipeline whose access policy or role assignment was removed, or whose address is no longer allowed by the vault firewall, and which keeps retrying. Fix the access or exclude the address.
  • Inventory and compliance scanners that probe every vault with an identity that has no data access.
  • The values of 20 requests and 10 minutes are starting points, not tested limits.

Key Vault Secret, Key or Certificate Backed Up

Detects a successful backup of a secret, a key or a certificate from an Azure key vault (operations SecretBackup, KeyBackup and CertificateBackup in the vault's AuditEvent resource log). The backup downloads the object with all its versions as an encrypted blob that can be restored into another key vault in the same subscription and Azure geography, where the copy is independent of the original. Microsoft recommends monitoring these operations and alerting on unexpected activity, because there is no way to revoke a backup blob once it exists. Key Vault has no operation that backs up a whole vault, so a series of these records from one address means that somebody is copying the vault object by object. The log exists only where a diagnostic setting on the vault exports the AuditEvent category.

LevelMediumStatusExperimentalLog sourceAzure Key Vault resource logs (AuditEvent)
title: Key Vault Secret, Key or Certificate Backed Up
id: d3c6c70f-e90d-4fd8-8d60-3734e9f10012
status: experimental
description: >-
  Detects a successful backup of a secret, a key or a certificate from an
  Azure key vault (operations SecretBackup, KeyBackup and CertificateBackup in
  the vault's AuditEvent resource log). The backup downloads the object with
  all its versions as an encrypted blob that can be restored into another key
  vault in the same subscription and Azure geography, where the copy is
  independent of the original. Microsoft recommends monitoring these
  operations and alerting on unexpected activity, because there is no way to
  revoke a backup blob once it exists. Key Vault has no operation that backs up
  a whole vault, so a series of these records from one address means that
  somebody is copying the vault object by object. The log exists only where a
  diagnostic setting on the vault exports the AuditEvent category.
references:
  - https://attack.mitre.org/techniques/T1555/006/
  - https://learn.microsoft.com/en-us/azure/key-vault/general/backup
  - https://learn.microsoft.com/en-us/azure/key-vault/general/logging
author: Forestall
date: 2026-11-18
tags:
  - attack.credential-access
  - attack.t1555.006
logsource:
  product: azure
  service: keyvault
detection:
  selection:
    OperationName:
      - SecretBackup
      - KeyBackup
      - CertificateBackup
    ResultType: Success
  condition: selection
falsepositives:
  - Planned backups of objects that cannot be created again, and migrations of objects between vaults or regions. Microsoft advises backing up only with a business justification, so these should be rare. Exclude the address of a known backup job in a filter_ selection by CallerIpAddress.
level: medium

False positives

  • Planned backups of objects that cannot be created again, and migrations of objects between vaults or regions. Microsoft advises backing up only with a business justification, so these should be rare. Exclude the address of a known backup job in a filter_ selection by CallerIpAddress.

Detection logic

Patterns that need more than one event, a baseline or knowledge of your own environment. Each is a hypothesis to build and tune in your SIEM, not a ready rule.

Secrets listed and then read one by one from the same address

Detection hypothesisSequence of events

An application asks for the secrets it knows by name. Somebody who does not know what a vault holds lists it first and then reads what the listing returned. Microsoft Defender for Key Vault describes a secret listing followed by secret reads as typically associated with secret dumping. The order of the two operations from one address is the signal, which a count of reads alone does not show.

  1. In the Key Vault resource log, find OperationName SecretList with ResultType Success. Keep CallerIpAddress and ResourceId (the vault).
  2. Within 10 minutes, count the SecretGet records with the same CallerIpAddress and ResourceId and a different Id each. Five or more different secrets is a starting point.
  3. Read the identity claims of the matching records to name the caller, and compare it with the identities that normally read this vault.

EventsSecretListSecretGet

LimitsConfiguration providers that list a vault and load every secret at application start do exactly this, so their addresses have to be excluded. A person who browses the secrets in the Azure portal also lists first. The counts and the window are starting points. An attacker who already knows the secret names does not list.

Secrets read from an address that never read this vault before

Detection hypothesisNeeds a baseline

The clients of a vault are few and stable, such as the outbound addresses of an application, a pipeline and a few administrators. A successful secret read from an address that is new for that vault is the simplest sign that a credential for the vault is being used from somewhere else.

  1. From the Key Vault resource log of the last 30 days, build the list of CallerIpAddress values per ResourceId for SecretGet with ResultType Success.
  2. For each new successful SecretGet, check whether its CallerIpAddress is in the list for that vault.
  3. Review reads from a new address, and give priority to vaults that hold production credentials and to addresses that are new for several vaults on the same day.

EventsSecretGet

LimitsApplications on platforms with changing outbound addresses, new deployments and administrators who work from new locations are new addresses as well. The 30 days are a starting point. An attacker who reads the vault from inside the application's own environment uses a known address and is not seen.

Triage

  • Identify the vault (the resource ID of the record) and the caller. The rules group by the caller's IP address. Open one of the counted records and read its identity claims to see which user or application made the requests.
  • Check whether this address and identity read from this vault routinely. An application that loads all its secrets at startup, a backup job and a secret scanner produce the same burst.
  • List the object identifiers that were read and what each secret gives access to. Secrets that the caller's application never used before weigh most.
  • Look at what happened on the vault before the reads, such as an access policy change, a data role assignment or a change of the vault's network rules, and who made it.
  • For a backup operation, find out where the blob could have been restored. Look for SecretRestore, KeyRestore or CertificateRestore in the logs of the other vaults in the subscription.

Response

Once the attack is confirmed:

  1. Remove the caller's access to the vault. Delete the access policy or the role assignment that gives it data access, revoke the sessions of the identity and rotate its credentials.
  2. Rotate every secret that was read, at its source (the database, the storage account, the application registration), and store the new value. Start with credentials of privileged identities and production systems.
  3. Reissue certificates whose private key was exportable and was read as a secret, and revoke the old ones.
  4. For keys that were backed up without authorization, follow Microsoft's sequence. Create a new key in a vault with restricted access, move the dependent services to it and only then disable the old key, because disabling it first takes those services offline.
  5. Review the sign-in and activity logs of the systems whose credentials were in the vault for use of those credentials after the reads, and continue the response there.
  6. If the vault had no diagnostic setting at the time, there is no record of what was read. Treat every object that the caller's permissions covered as read.

Mitigation

  • Create a diagnostic setting on every vault that sends the AuditEvent category to your SIEM. Without it there is no record of which secrets were read.
  • Use the Azure RBAC permission model and grant Key Vault Secrets User to the application identity that needs the secrets. Keep Key Vault Administrator, Key Vault Secrets Officer and the backup permissions to a few identities, assigned through Privileged Identity Management.
  • Use one key vault per application, region and environment, as Microsoft recommends, so that one compromised identity reaches the secrets of one application and not of all.
  • Disable public network access and use private endpoints, or enable the Key Vault firewall, so that a stolen token cannot read the vault from the internet.
  • Limit the roles that can change the vault's access configuration (Owner, User Access Administrator, and Contributor or Key Vault Contributor on vaults with access policies).
  • Enable Microsoft Defender for Key Vault, which alerts on unusual access patterns that a fixed threshold cannot express.

Frequently asked questions

Does the Azure Activity log show that secrets were read from a key vault?

No. Reading a secret is a data-plane request to the vault. The Activity log records control-plane operations such as changing an access policy or deleting the vault. Secret, key and certificate operations are written to the vault's resource log, category AuditEvent, and only when a diagnostic setting on that vault exports the category.

Why do the rules group by IP address and not by the user or application?

The Key Vault log carries the caller's identity as a set of token claims in one field, and the structure of that field is not mapped as a searchable field in this library. The caller's IP address and the vault are plain fields, so the threshold rules count per address and vault. Read the identity claims in the matching records during triage to name the caller.

Will an application that loads its configuration from the vault trigger the threshold rule?

It can. An application that reads all its secrets when it starts produces many different SecretGet requests within seconds, from the same address, at every restart. Exclude the addresses of such applications once they are confirmed, or raise the count above what they read. The rule is most useful on vaults whose normal clients read one or two secrets.

Can a Key Vault backup file be used outside Azure?

Not directly. Microsoft documents that the backup is an encrypted blob that cannot be decrypted outside Azure and must be restored into a key vault in the same subscription and Azure geography. Inside that boundary a restored copy is independent of the original, and Microsoft states that there is no way to revoke a backup blob or a restored key.

References

We respect your privacy

We use cookies to keep this site secure and working properly. With your permission, we also use optional cookies to understand usage and improve the experience. Cookie Policy

You can change your choice at any time.