Three identity stories converged this morning that share a single through-line: the trust boundary has moved upstream, and most organizations are not watching it. An OAuth federation flaw let attackers bypass MFA entirely. A CVSS 10.0 VPN zero-day turned the perimeter device into the initial access vector. And 153 million stolen driver’s licenses undercut the document-based identity verification ecosystem. Each story is a different failure mode of the same problem — trusting an identity source that should not have been trusted.

In the News

Dropbox Accounts Breached via Lenovo SSO — OAuth Trust-Chain Failure

Attackers exploited a broken email-verification flow in Lenovo’s identity provider to register accounts using email addresses they did not control. Those fraudulent Lenovo identities were then federated via SSO into Dropbox, granting access to accounts tied to the spoofed email domains. BleepingComputer reports the attack chain bypassed MFA entirely — the authentication challenge happened at Lenovo’s IdP, which the attacker controlled from the start.

This is a textbook federated identity trust failure. The attacker never touched a password, never triggered an MFA prompt on the victim’s device, and never needed to phish anyone. They operated upstream of every security control by exploiting the weakest link in the trust chain: the identity provider’s verification of who the user actually is. MITRE ATT&CK maps this to T1078.004 (Valid Accounts: Cloud Accounts) and T1550.001 (Use Alternate Authentication Material: Application Access Token).

The operational takeaway is uncomfortable: SSO reduces password sprawl, but it concentrates trust. If the upstream IdP’s enrollment or verification process is compromised, every downstream relying party inherits that compromise. Organizations with complex federation topologies — multiple IdPs, B2B SAML trusts, OAuth app integrations — need to treat IdP verification integrity as a critical control, not an assumed baseline. Continuous session validation and ITDR tools that flag anomalous token issuance from federated sources are the detection layer that catches this before data leaves.

What defenders should do: Audit all federated SSO trust relationships. Identify which IdPs your organization trusts and evaluate their enrollment verification controls. Deploy ITDR tooling that monitors for anomalous federated token issuance — new IdP sources, accounts that appear via federation paths that have never been active, and impossible-travel patterns on federated sessions. Enforce phishing-resistant passwordless authentication (FIDO2) at the resource layer so that even a compromised upstream IdP cannot bypass the final authentication challenge.

SonicWall SMA 1000 Zero-Days Chained for Pre-Auth RCE

Two zero-day vulnerabilities in SonicWall’s SMA 1000 SSL VPN gateway — a CVSS 10.0 server-side request forgery (CVE-2026-83548) chained with an authentication bypass — give attackers pre-authentication remote code execution on the appliance. The Hacker News reports that SonicWall discovered the flaws internally but exploitation was already underway at the time of disclosure.

The attack chain requires no credentials. The SSRF allows the attacker to reach internal services on the appliance, and the auth bypass removes the remaining barrier to code execution. Once an attacker has RCE on the VPN gateway, they have a foothold inside the network perimeter with access to authentication material, session tokens, and the gateway’s network position. This is T1190 (Exploit Public-Facing Application) leading directly into T1078 (Valid Accounts) via harvested VPN credentials.

This is not the first time a VPN appliance has become the breach point — Ivanti, Fortinet, and Citrix have all faced similar zero-day chains in the past 18 months. The pattern is consistent: internet-facing VPN concentrators are high-value targets, and the attack surface they present is structurally difficult to defend when the vulnerability is pre-auth.

What defenders should do: Patch SonicWall SMA 1000 immediately or take it offline. Review VPN gateway logs for anomalous SSRF patterns and unexpected outbound connections from the appliance. Long-term, evaluate ZTNA/SSE architectures that eliminate the internet-facing appliance entirely.

153M+ Driver’s Licenses for Sale — FBI Probes Identity Verification Firm

Krebs on Security reports that a dark-web service is offering scans of over 153 million U.S. and Canadian driver’s licenses, apparently sourced from a Louisiana-based identity verification platform. The FBI’s New Orleans field office has opened an investigation. The firm has not been publicly named in the FBI probe.

The scale of this breach is significant not because of the number alone, but because of what the data enables. Driver’s license scans are the primary document used in KYC (Know Your Customer) identity verification across financial services, telecommunications, and healthcare. With 153 million real document images available, the barrier to constructing synthetic identities — combining a real document with fabricated supporting information — drops to near zero. Every organization that relies on document-based identity proofing during account enrollment now faces a degraded trust baseline.

What defenders should do: Organizations relying on document-based identity verification for enrollment should layer additional proofing signals — liveness detection, device-bound credential enrollment, and behavioral analytics during onboarding. Phishing-resistant authentication (FIDO2) bound to a physical device cannot be replicated from a stolen document image.

Today’s Deep Dive — AI Platforms as Credential Vaults: The Langflow and METR Incidents

Two incidents this week illustrate a pattern that identity and security teams need to internalize: AI platforms are becoming the highest-density targets for non-human identity (NHI) credential theft.

CVE-2026-0768 — Langflow RCE for API Key Harvesting. Langflow, a Python-based framework for building AI applications, has a critical unauthenticated RCE vulnerability (CVSS 9.8) that attackers are actively exploiting to steal OpenAI and AWS credentials stored in environment variables. The attack is straightforward: exploit the app-layer RCE, dump environment variables, exfiltrate API keys. No lateral movement required — the credentials are on the host. MITRE ATT&CK: T1552.001 (Unsecured Credentials: Credentials in Files).

METR API Key Theft — $600K in AI Compute Credits. Separately, METR, an AI safety research nonprofit, had API keys stolen and watched approximately $600,000 in cloud AI compute credits get burned by the attacker. This is the financial dimension of NHI credential theft that gets CFO attention: stolen API keys do not just access data — they access compute budgets.

The common thread is that AI development environments concentrate NHI credentials — API keys for cloud providers, model endpoints, vector databases, and orchestration services — on hosts and in configuration files that were never designed as credential stores. Secrets management (HashiCorp Vault, AWS Secrets Manager, Azure Key Vault) exists precisely to solve this, but adoption in AI/ML development environments lags significantly behind traditional application development. GitGuardian’s recent analysis reinforces this: EDR and XDR tools detect malicious behavior on the endpoint but do not inventory which valid credentials are sitting exposed right now.

What defenders should do: Audit AI development environments for hardcoded or environment-variable-stored credentials. Migrate API keys to secrets management platforms with automatic rotation. Deploy secrets-scanning tools on endpoints and in CI/CD pipelines. Implement workload microsegmentation to limit what a compromised AI application host can reach even after credential theft.

Detection Spotlight

Detecting anomalous federated SSO token usage — the pattern from the Dropbox/Lenovo breach — requires monitoring for federation sources that are new or rarely used. The following KQL query for Microsoft Sentinel surfaces sign-ins that arrive via a federated IdP not seen in the previous 30 days for that user:

let lookback = 30d;
let known_federations = SigninLogs
    | where TimeGenerated between (ago(lookback) .. ago(1d))
    | where ResultType == 0
    | where isnotempty(HomeTenantId) and HomeTenantId != ResourceTenantId
    | distinct UserPrincipalName, HomeTenantId;
SigninLogs
    | where TimeGenerated > ago(1d)
    | where ResultType == 0
    | where isnotempty(HomeTenantId) and HomeTenantId != ResourceTenantId
    | join kind=leftanti known_federations on UserPrincipalName, HomeTenantId
    | project TimeGenerated, UserPrincipalName, HomeTenantId, AppDisplayName, IPAddress, Location

This query flags sign-ins from federated tenants that have never authenticated a given user before. False positives include legitimate new B2B collaborations — tune by excluding known partner tenant IDs. High-fidelity in environments with stable federation topologies.

Defender Action Items

  • Patch SonicWall SMA 1000 immediately or take offline — CVE-2026-83548 is pre-auth RCE with active exploitation
  • Patch Langflow (CVE-2026-0768) or isolate instances — attackers are harvesting API keys from environment variables
  • Patch JFrog Artifactory (CVE-2026-82329) — auth bypass allows admin token minting, supply-chain risk
  • Audit all federated SSO trust relationships — identify which upstream IdPs your organization trusts and evaluate their enrollment verification
  • Deploy ITDR tooling that monitors for anomalous federated token issuance and new federation sources
  • Audit AI development environments for hardcoded credentials and migrate to secrets management

References

Subscribe to The Identity Brief

Get The Identity Brief in your inbox (Mon/Wed/Fri) — Human, machine, and AI identity security — NHI, ITDR, and the IAM market.