7.1

CVSS 3.1 base score, per SonicWall; the attacker needs a valid password, not nothing

0

workarounds: the fixed firmware is the only control SonicWall offers

7

CVEs fixed in the one advisory; this is the only one marked as potentially exploited

20 Nov

federal deadline CISA set to apply the fix, 14 days after listing

The MFA was never asked

A login is only as strong as the weakest method the appliance still honours. SonicWall confirmed CVE-2026-19868 on 6 Nov 2026, and CISA listed it as exploited the same day: on a SonicOS firewall where the SSL-VPN uses SAML single sign-on, a user with valid credentials for a second authentication method could reach a VPN session without ever going through SAML, and so without the multi-factor prompt that SAML was enforcing. The MFA was not broken. It was skipped, because the appliance kept a quieter door open beside the one everyone was watching.

That is the shape of the finding, and it is the shape of most identity failures we see. An organisation buys an identity provider, turns on SAML for the VPN, enforces MFA in the provider's conditional access policy, and reasonably concludes the VPN is behind MFA. The policy is real. It just lives on the identity provider, and it can only act on a login the identity provider is asked to handle. The appliance in front of it decides which logins to send there, and if it will also accept a username and password on its own, the strongest policy in the estate never runs.

We have written this argument before, in plainer terms, about the ways past MFA we use on most engagements. This is the same lesson with a vendor advisory attached. The number on the CVE is 7.1, which reads as moderate. The reach is not moderate. A remote access gateway is the front door to the internal network, and a way in that your detection cannot distinguish from a normal employee is worth more to an operator than most remote code execution bugs.

What is SonicWall CVE-2026-19868?

SonicWall published advisory SNWLID-2026-0014 on 6 Nov 2026. It covers seven SonicOS flaws; CVE-2026-19868 is the one the vendor flags for exploitation. SonicWall describes it as an authentication enforcement vulnerability in the SonicOS SSL-VPN service that may allow a user with valid credentials for an alternate authentication method to bypass SAML authentication controls, and notes that successful exploitation may also bypass SAML-enforced MFA. The advisory rates it CVSS 7.1 with the vector AV:N/AC:H/PR:L/UI:N/S:C/C:H/I:L/A:N, and classifies it as CWE-308, use of single-factor authentication. The vector carries two facts worth reading: the attacker needs some privilege already (PR:L, a valid credential for another method), and the scope is changed (S:C, the weakness lets them reach beyond the component that failed).

The condition that decides whether you are exposed is narrow and specific. SonicWall states the issue only affects the SSL-VPN if SAML is enabled for SSL-VPN logins. A firewall whose VPN never used SAML is not in scope for this one. A firewall that moved its VPN behind SAML single sign-on, which is the modern, recommended configuration, is exactly the one at risk, because the bug is in how the appliance enforces that SAML step against accounts that can also authenticate another way.

On exploitation, we stay with the sources' own words. SonicWall says its product security team has investigated a report indicating potential exploitation of this vulnerability and urges customers to upgrade. SonicWall has not shared details of the suspected attacks, and it is not established whether the flaw was used as a zero-day before the fix. CISA is firmer: it added CVE-2026-19868 to the Known Exploited Vulnerabilities catalog on 6 Nov based on evidence of active exploitation, and set a federal remediation deadline of 20 Nov. So the honest reading is that the vendor reports a potential exploitation report while the federal catalogue treats exploitation as established.

CVE-2026-19868, SonicOS SSL-VPN High
  • Product //

    SonicWall SonicOS, the SSL-VPN service, Gen7 and Gen8 firewalls and NSv virtual appliances
  • Condition //

    only when SAML is enabled for SSL-VPN logins
  • Affected //

    Gen7 7.3.3-7015 and earlier, Gen8 8.2.2-8015 and earlier
  • Fixed in //

    Gen7 7.3.4-7013, Gen8 8.2.3-8015
  • CVSS //

    7.1 (3.1, SonicWall), CWE-308 use of single-factor authentication
  • Workaround //

    none; the fixed firmware is the only control
  • Exploited //

    CISA KEV lists it as exploited; SonicWall reports a potential exploitation report

The other six flaws in the advisory are not marked as exploited, and we mention them only so an operator patches with the full picture. Two are rated 8.8 and sit in the SonicOS management interface: a permissive cross-domain policy issue and a session validation issue, each able to let a remote unauthenticated attacker perform unauthorized operations under specific conditions. Three are pre-authentication buffer overflows, in the SSL-VPN service and the TACACS+ authentication client, that can cause an application level denial of service. One is weak password hashing on virtual appliances, reachable only with existing local access. The single firmware upgrade closes all seven.

How a password alone becomes a VPN session

The chain is short, and we describe it at the level of what each step achieves rather than how to perform it. There is nothing here that helps an attacker who does not already have the advisory, and the one thing that matters for a defender is where the MFA policy stops being able to act.

An operator starts with something they very often already have: a valid username and password for an employee. That is the ordinary output of a phishing page, an infostealer log, a credential stuffing hit, or a password reused from a breach. On a correctly configured SAML VPN that credential alone should be worth little, because the login is supposed to be handed to the identity provider, which applies conditional access and demands a second factor. The bug is that the appliance will honour that credential through an alternate authentication method and mint the session itself, without handing the login to the identity provider at all. The second factor is never requested, because the system that would request it is never consulted. What the operator ends up with is a VPN session that looks, from the inside, like any other remote employee.

Figure 01

A valid password, and no prompt

  1. 01

    Hold //

    The operator has a valid username and password for one employee, the common currency of phishing and infostealer logs.

  2. 02

    Reach //

    They reach the SonicWall SSL-VPN portal, which the organisation put behind SAML single sign-on with MFA.

  3. 03

    Skip //

    The appliance accepts the credential through an alternate method and never hands the login to the identity provider.

  4. 04

    Silent //

    The SAML step and the MFA prompt behind it do not run, because the system that enforces them is not asked.

  5. 05

    In //

    A VPN session is established that looks like a normal remote employee from inside the network.

Result //

The MFA was not defeated. It was never invoked.

The policy that would have stopped this lives on a system the login never reached.

The limits matter and we state them plainly. The operator needs a valid credential; this is not an unauthenticated path to the network, and the CVSS vector says as much with PR:L. It needs the SSL-VPN to be configured for SAML, which is the condition SonicWall names. What it does not need is the second factor, the thing the organisation is relying on to make a stolen password survivable. For a remote access gateway, removing the second factor is most of the battle, because a password that should have been contained by MFA now reaches all the way to a tunnel into the network.

Why SSO on the VPN did not mean SSO only

The instructive part is not the specific bug. It is why a reasonable, well-run configuration was exposed by it. Three ordinary decisions line up.

The first is that enabling a strong method rarely disables the weak ones. Turning on SAML for the VPN added a good path. It did not necessarily remove the appliance's own local or directory-backed authentication, which stayed available as an alternate method. Most products keep several authentication methods live at once so that a provider outage does not lock everyone out, and so that service accounts and break-glass access keep working. Every method left enabled is a path an attacker may choose, and the attacker always chooses the weakest one you still accept.

The second is that the enforcement and the policy live in different places. The MFA requirement is expressed in the identity provider's conditional access rules. Those rules are evaluated only when the identity provider is asked to evaluate a login. The decision about whether to ask it belongs to the appliance. When the two disagree, when the appliance accepts a login the provider never sees, the policy is not weakened, it is bypassed, and nothing in the provider's logs will show a denied attempt, because there was no attempt to deny. We made the neighbouring argument about Palo Alto VPNs, where a forged cookie logged attackers in as any user and the rated severity understated the reach. The pattern is the same: the control is real, and it is in the wrong place to see the attack.

The third is that this failure mode is invisible to the usual checks. A configuration review confirms SAML is enabled and MFA is enforced, and both are true. A penetration test that assumes the VPN is behind MFA may never try a plain credential against it. The gap only appears when someone asks the harder question, which is not whether the strong method is turned on, but whether any weaker method is still accepted and what it is worth. That question is the whole of our identity work, and it is why we treat every remaining authentication method as a control to be tested rather than a convenience to be trusted.

How to find VPN logins that skipped the identity provider

The clean signal for this class of bypass is a disagreement between two logs that should agree. When SAML single sign-on is working, every SSL-VPN session on the firewall has a matching sign-in event on the identity provider: the firewall accepted the user because the provider vouched for them. A login that bypasses SAML breaks that pairing. The firewall records an accepted VPN session, and the identity provider has no sign-in for that user at that moment, because it was never consulted.

So the hunt is a reconciliation, not a signature. Pull the SSL-VPN authentication events from the SonicOS logs, pull the sign-in events from the identity provider for the same users and the same window, and look for VPN sessions with no provider sign-in behind them. On a healthy SAML deployment that set is empty. Anything in it is either a misconfiguration worth fixing or a login that went around the provider, and both deserve a ticket. The query below sketches the identity-provider half in KQL, against Entra ID sign-in logs, as a place to start; the reconciliation itself is done by joining it to the firewall's own VPN log, which is specific to your collection.

Figure 02

VPN users with no interactive SSO sign-in in the window

KQL

// Entra ID half of the reconciliation. Join the result to your
// SonicWall SSL-VPN accepted-login events (same user, same time window)
// and alert on VPN sessions that have no matching sign-in here.
let window = 1h;
let vpnApp = "SonicWall SSLVPN"; // the SAML app registration name in your tenant
SigninLogs
| where TimeGenerated > ago(7d)
| where AppDisplayName == vpnApp
| project TimeGenerated, UserPrincipalName, IPAddress,
          ResultType, AuthenticationRequirement,
          mfa = tostring(parse_json(AuthenticationDetails))
| summarize ssoSignIns = count() by UserPrincipalName, bin(TimeGenerated, window)
// Left-join your firewall VPN logins to this; a VPN login with
// ssoSignIns == 0 for that user and window is the trace to investigate.
A starting point to tune against your own telemetry, not a production rule. Confirm the app name and your firewall log schema before relying on it.

A single orphan login is not proof of attack on its own, because clock skew, cached sessions and service accounts produce false pairs. The value is the direction it points. On a VPN that is supposed to be SAML only, the set of sessions with no provider sign-in should be close to empty, and it is the exact set this bug produces. If you find one, treat the account as potentially compromised: rotate its credential, check what the session reached inside the network, and review your other remote-access appliances for the same split between a strong method you enabled and a weak one you left on. Our guide to responding to a stolen session or token covers the containment steps once a login like this turns up.

How we test remote access on an engagement

Remote access gateways are the first thing we look at on an external test, because they are internet-facing by definition and they are the shortest path from a stolen password to the inside. The question we ask is not whether the VPN supports MFA. It is which authentication methods the portal will actually accept, and what each of them is worth on its own. A portal that offers SAML single sign-on and, quietly, still takes a local username and password has two doors, and we test both.

Enumerating the methods a login surface honours is ordinary external work, and we keep it to that: we establish what the portal will accept and we stop at the point where a method proves weaker than the organisation assumes, without taking a client credential any further than needed to demonstrate the gap. The finding is that the gateway accepts a path that skips the MFA the organisation is counting on, and that finding stands whether or not anyone walks through it. Our external recon method maps the remote-access surface an organisation exposes from its own ranges, and our checks on remote administration services and stolen token response cover how we test what those services honour and how a defender should react when a session appears that no second factor gated. The password checker on our tools page is a small piece of the same point: a credential's strength only matters if every method that accepts it is one you meant to leave open.

The broader habit this bug rewards is treating MFA as a property of a login path, not of a system. An organisation does not have MFA; a specific way of logging in does, and only while every other way is either removed or just as strong. When we report a VPN finding, the recommendation is rarely only patch. It is also enumerate the methods this gateway accepts and close or harden the ones you did not mean to keep, because the next bug in this class will land on whichever one you forgot.

What is still coming

Expect more detail on the exploitation, and expect it slowly. SonicWall has not described the suspected attacks, and CISA's listing states exploitation without narrating it. SonicWall SSL-VPN appliances have been a repeated target through 2026, so a gateway bug that removes the second factor is the kind of thing that moves from a quiet report to broad scanning quickly once the condition is understood. The practical takeaway does not wait for that detail: the fixed firmware is available, there is no workaround, and the exposure is live now.

Expect the same class on other vendors. The weakness here is not unique to SonicOS. Any gateway that supports several authentication methods at once, and enforces its strongest one through an external identity provider, can have the same split between the method it advertises and a method it still accepts. We have now written about it on SonicWall, on Palo Alto's forged VPN cookie, and around the edges of the Microsoft 365 token replay that survived a password reset. The through-line is that identity controls fail at the seams between systems, where one component assumes another did the checking.

Expect your own inventory to be the hard part. The fix is a firmware upgrade, which is easy to describe and tedious to complete across a fleet of firewalls, some of them at sites nobody logs into for months. The detection is a log reconciliation, which most organisations have never set up because, until a bug like this, there was no reason to doubt that a SAML VPN login implied a SAML sign-in. Building that reconciliation once is worth more than this single patch, because it is how you will catch the next method that turns out to skip the one you were watching.

What to do today

  1. Upgrade affected SonicWall firewalls now; there is no workaround. Gen7 to 7.3.4-7013, Gen8 to 8.2.3-8015. The fixed firmware is the only control SonicWall offers, and CISA set a federal deadline of 20 Nov.

  2. List every firewall whose SSL-VPN uses SAML single sign-on. Those are the exposed ones. A VPN that never used SAML is out of scope for this flaw, but confirm that rather than assume it.

  3. Audit which authentication methods each VPN portal still accepts. Enabling SAML did not remove the local or directory-backed methods beside it. Close or harden the ones you did not mean to keep.

  4. Reconcile firewall VPN logins against your identity provider sign-ins. A VPN session with no matching SSO sign-in is the trace this bypass leaves. The KQL above is a starting point for the provider half.

  5. Treat the other six advisory flaws as part of the same upgrade. Two management-interface bugs are rated 8.8; the single firmware release closes all seven.