Posted in Identity and Access Management

Your MFA rollout doesn't cover the identities attackers are now signing in as. This week Todyl published an…

Craig M. — Chief Product Officer @ Cyberr®

Your MFA rollout doesn't cover the identities attackers are now signing in as.

This week Todyl published an investigation into a single operator who used stolen Entra ID app secrets and Azure DevOps identities to reach 27 organisations, many of them MSP clients. No phishing. No MFA prompts. Just client credentials signing in as the workload.

The pattern per credential was fast: Azure Resource Manager token, then Key Vault, then Microsoft Graph and Storage. Wildcard searches across tenants followed, then SharePoint downloads at roughly one file per second, targeting password lists, VPN profiles and API keys.

The gap: service principals don't do MFA, user-scoped Conditional Access doesn't apply to them, and most detections watch interactive user sign-ins. Several of the stolen secrets had only ever been used from staff devices and office networks. Nobody was baselining that.

What I'd check:

→ Query AADServicePrincipalSignInLogs for invalid or expired secret failures (AADSTS7000215 / 7000222) followed by a success for the same app.

→ Inventory app registrations still using client secrets. Move to managed identities or federation, and use app management policies to block new secrets.

→ Apply Conditional Access for workload identities to single-tenant apps that should only run from known locations.

→ Lock Key Vault and Storage to private endpoints or selected networks.

In Heelr by Cyberr, the Heelr Compliance module continuously checks baselines like MFA enforcement for every Microsoft 365 user. That baseline still matters. But the harder question sits one layer down.

How many app secrets in your tenant are older than the people who created them?

#EntraID #IdentitySecurity #CloudSecurity

4 reactions · 3 comments

Topics: #identitysecurity, #cloudsecurity, #entraid

Comments

Schomer: You hit the nail on the head, Craig. The "harder question sits one layer down" because standard compliance modules and behavioral baselining are fundamentally reactive—they only look at session telemetry after an identity token has already been accepted and parsed by the host runtime daemon. If a service principal authenticates using a stolen client secret, traditional systems grant ambient trust.

Victor Font: Craig, this exposes a governance blind spot: organizations rigorously manage human identities while allowing workload credentials to operate with less oversight. The critical question isn't simply how old those secrets are, but who owns them, whether their privileges remain justified, and what evidence demonstrates continuous oversight. An identity without accountability is an unmanaged risk, huma

Schomer: Victor, you’ve hit on the exact structural disconnect. Traditional IT governance relies heavily on "point-in-time" accountability—auditing ownership or checking a box during a quarterly review—but that paper trail offers zero defense against machine-speed credential abuse. An identity without real-time, continuous validation is just a liability waiting to be exploited. This is precisely where gove

Read and join the discussion on Cyberr