Critical Infrastructure

Cisco SD-WAN Zero-Day CVE-2026-20245 Gave Attackers Root Inside a Telecom Fabric

An unknown actor ran Cisco Catalyst SD-WAN as a zero-day for two-plus months, escalating a stolen admin account to full root on a telecom provider. Mandiant traced a malicious CSV upload, a hidden 'troot' account, and disciplined anti-forensics.

Matt Lucas  |  June 28, 2026  |  6 min
Detected by CaverLive detection for 2 CVEs in the RedEye Intel Feed →
Editorial illustration: woven SD-WAN network fabric converging on root-access keys with red corruption radiating out
CVEs in this postCVE-2026-12569CVE-2026-20127CVE-2026-20182CVE-2026-20245Live detections →All RedEye CVEs →
7.8
CVSS score
2+ months
Zero-day before disclosure
3
CVEs chained
root
Privilege achieved
TL;DR
  • What: An unknown threat actor exploited Cisco Catalyst SD-WAN flaw CVE-2026-20245 as a zero-day, uploading a malicious CSV (evil_tenant.csv) to escalate a netadmin account to full root on a communications service provider.
  • Impact: The attacker gained root shell via a hidden 'troot' account, exfiltrated the SD-WAN fabric configuration, and then erased nearly all evidence using restore-and-revert anti-forensics.
  • Fix / mitigation: Apply Cisco's patches for CVE-2026-20245, CVE-2026-20127, and CVE-2026-20182, rotate all admin credentials and certificates, and rebuild trust on devices showing rogue peering.
  • Who's at risk: Communications service providers and any organization running Cisco Catalyst SD-WAN controllers and edge devices at network chokepoints.

An unknown threat actor ran inside a communications service provider's Cisco Catalyst SD-WAN fabric for at least two months before anyone knew the bug existed. According to Mandiant, the actor weaponized CVE-2026-20245 (CVSS 7.8) as a zero-day, turned a compromised admin account into full root, stole the fabric configuration, then deleted nearly every artifact behind them. Cisco confirmed active exploitation earlier this month.

This is the edge-device problem in one incident: a high-value box that sees internal traffic across the fabric, has no native EDR, and lacks the telemetry defenders need to reconstruct what happened. The attacker exploited all three conditions deliberately.

The intrusion chain

Mandiant identified two distinct windows of unauthorized activity, and it is still unclear whether the same actor was behind both. The first ran from late 2025 into January 2026; the second hit in March 2026. Both began with rogue peering connections into the SD-WAN controllers.

In the first wave, the rogue peering likely abused one of two authentication bypass flaws, CVE-2026-20127 or CVE-2026-20182, both undisclosed zero-days at the time. In March, the target device had already been patched against CVE-2026-20127, and Cisco confirmed the connections did not use CVE-2026-20182. That points to stolen certificates from a prior breach of the same device, the attacker reusing trust they had already harvested rather than burning a fresh exploit.

From netadmin to root via a CSV file

CVE-2026-20245 requires an authenticated local attacker with netadmin privileges, then exploits the device's insufficient validation of a user-supplied file. The actor changed the default admin credentials, then uploaded a crafted CSV named evil_tenant.csv. That single file triggered the privilege escalation.

With root, the attacker created a hidden account called 'troot' written directly into /etc/passwd and /etc/shadow, giving them a full root shell that did not depend on the original compromised account. The crafted-file pattern is exactly why edge appliances keep falling: a parser that trusts operator input becomes a code-execution primitive.

Anti-forensics by design

After exfiltrating the SD-WAN fabric configuration, the actor changed the admin password back to its original value so a logging-in administrator would notice nothing. They then deleted every file they touched, reverted their configuration changes, and ran a validation script to confirm their indicators were gone. On a device with limited logging, that combination can erase the intrusion entirely.

Why edge devices keep losing

Google Threat Intelligence Group framed this as a continuing trend: advanced adversaries primarily target network devices and systems that do not support EDR. Mandiant CTO Charles Carmakal made the same point publicly. An SD-WAN controller is an ideal foothold because it sits at a traffic chokepoint, offers persistent visibility into internal flows across the fabric, and cannot easily be watched the way an endpoint can.

The selective delete-and-restore tradecraft here was not opportunistic. It was the work of an actor who understood that the device's weak telemetry was their best ally, and who invested in staying invisible rather than moving fast.

What to do now

The takeaway

Three zero-days, stolen certificates, root via a CSV file, and a cleanup routine that validated its own success. If your network appliances cannot be monitored like endpoints, assume they are the soft target, and build detection around the traffic and configuration state they cannot delete.

The victim was a communications service provider, but the lesson generalizes to anyone running unmonitored network gear at a chokepoint. The exploit was not loud or clever in execution; it was patient, and it counted on no one looking. That is the part worth fixing first.

Questions about your exposure?

RedEye Security provides assessments for organizations that need to understand their real risk.

Talk to us