The Patch Is Not the Recovery Plan
A patch can close an authentication bypass without telling you whether the attacker already used it.
That distinction is the important part of CVE-2026-104935, a newly exploited Ivanti Connect Secure vulnerability added to CISA’s Known Exploited Vulnerabilities Catalog on October 13, 2026.
The flaw is an authentication bypass by spoofing. NVD describes a remote, unauthenticated attacker executing arbitrary commands. CISA says threat actors were exploiting it for credential theft and data exfiltration.
The operator lesson is not simply “patch Ivanti.” It is that an appliance at the edge of the network is both a software product and an evidence boundary. Updating the first without preserving the second can leave the real incident unanswered.
What changed
CISA says the affected versions include Ivanti Connect Secure 25.1.1.1 and prior, 25.1.2.2 and prior, and 22.8R2.4 and prior. NVD records the fixed versions as 25.1.3.0, 25.1.3.1, 22.8R2.5, and 22.7R2.14, with 22.7R2.14 listed as unaffected.
The vulnerability is rated 8.6 High by Ivanti’s CNA record in NVD. The vector is network reachable, low complexity, and requires no privileges or user interaction.
CISA also gives affected operators a deadline of October 16, 2026 in the KEV record. That is a short window because the fact that exploitation is observed changes the decision from routine patching to active exposure management.
The version check is the easy part
The hard question is whether a vulnerable appliance was already used.
CISA recommends network and memory forensics. It explicitly warns that Ivanti’s Integrity Checker Tool may not detect exploitation of this vulnerability because threat actor commands execute in memory.
That warning changes the normal response sequence. A clean integrity check is not a clean bill of health. It is one signal from one inspection method, and this attack path may leave the most important activity outside what that method can see.
The recovery guidance is correspondingly stronger than “install the update”:
- Back up the appliance configuration.
- Perform a factory reset and upgrade the appliance. Rebuild cloud and virtual instances rather than treating them as trusted survivors.
- Restore the configuration from the backup.
- Reset all relevant passwords, including domain, administrator, and local accounts.
- Revoke and reissue stored certificates, API keys, and license server credentials.
The sequence is designed around a compromised control plane. Preserve what you need, discard the potentially compromised runtime, then rotate the secrets that may have crossed it.
The edge appliance is part of the identity system
It is tempting to classify a VPN or remote access gateway as a networking box. That is operationally misleading.
An appliance that authenticates users, terminates remote sessions, stores certificates, and connects to internal services sits on the path to identity and authority. If it is compromised, the blast radius is not limited to the appliance’s own filesystem.
Credentials can be stolen. Session context can be abused. Configuration can reveal internal routes and trusted integrations. Stored keys can turn a single appliance compromise into access to systems that never exposed the original bug.
This is why credential rotation belongs in the recovery procedure, not in a later cleanup ticket. The vulnerable appliance may have been the place where secrets were exposed, even if the attacker never modified a file that a post-patch checker can see.
What the news cycle gets wrong
The common security headline is “zero-day patched.” That framing compresses two separate questions into one:
- Is the vulnerable code still reachable?
- Should the organization trust the appliance and the secrets that passed through it?
The first is a remediation question. The second is an incident-response question.
CISA’s advisory makes the separation explicit. It names observed exploitation, recommends forensics, warns about in-memory commands, and calls for reset and reissuance actions after recovery. The interesting part is not that the patch exists. It is that the recommended response assumes the patch alone cannot establish trust.
The agent security connection
The same boundary failure appears in agent systems.
An agent runtime may report that a tool is up to date, that a policy file has the expected checksum, or that a container restarted cleanly. None of those facts proves that credentials, tokens, browser sessions, or tool outputs were not exposed before the check.
A trustworthy control loop needs two separate verifiers:
runtime verifier: is the component now in the expected state?
incident verifier: what could have happened before it got there?
The first supports safe operation. The second supports safe recovery.
This distinction matters for tool servers, browser gateways, MCP connectors, and local agent harnesses. If a component can reach privileged systems, a restart is not automatically a reset of trust. Secrets held by the component may need rotation. Logs may need preservation before cleanup. The operator needs an explicit boundary between evidence collection and remediation.
A practical response checklist
For an internet-facing appliance listed in KEV, the minimum useful checklist is broader than a package update:
| Question | Evidence to collect |
|---|---|
| Was the vulnerable version exposed? | Asset inventory, external scans, version history |
| Was exploitation possible? | Network telemetry, authentication logs, appliance and upstream logs |
| Could commands run only in memory? | Network and memory forensics, not just file integrity checks |
| What secrets did the appliance hold? | Configuration backup, certificate inventory, API key inventory |
| Can the runtime still be trusted? | Reset, rebuild, upgrade, and documented recovery procedure |
| What authority may have been reached? | Session records, downstream access logs, identity provider events |
The goal is not to collect every log forever. It is to make the trust decision explicit and evidence-based.
The operator consequence
Treat the KEV entry as a change in operating mode, not as a normal patch reminder.
Inventory every affected appliance. Identify internet exposure and dependent identity systems. Preserve the configuration and relevant evidence. Follow the vendor and CISA recovery procedure. Rotate credentials and reissue keys that crossed the boundary. Then verify both the fixed version and the downstream access that the appliance could authorize.
A version number tells you what code is running now. It does not tell you what ran before, what it touched, or what it disclosed.
The patch closes the door. Recovery decides whether you still trust the room.