August 2026 produced one genuine zero-day and eight vulnerabilities that were, in most cases, old news by the time anyone outside the vendor noticed them. That gap is the story of the month. A load balancer flaw from Progress, a CI server flaw from JetBrains, and a VPN appliance chain from SonicWall had all been publicly disclosed and patched before August began. None of that stopped a fresh wave of exploitation once August arrived, and in most cases it took a CISA Known Exploited Vulnerabilities catalog addition, not the original advisory, to get organisations moving at all.
A run of separate CVEs landed in the KEV catalog this month, spanning a load balancer, a CI server, an RMM platform, and an AI application framework. Cisco added a firewall VPN flaw to the pile, disclosed and under active attack within the same advisory. And N-able discovered that its first attempt at patching a critical N-central authentication bypass simply didn’t hold, handing attackers a second way into the same platform after the original fix shipped.
The one real zero-day of the month came from a state-sponsored actor, not a criminal crew. Lazarus Group had working exploit code for a Windows kernel driver flaw months before Microsoft patched it on August’s Patch Tuesday, using the access to deploy a rootkit built specifically to make its intrusions invisible to the security software running on the same machine. Everything else this month came down to organisations running software with a known, patchable hole in it for weeks, sometimes months, longer than they should have.
What We'll Cover
Windows Ancillary Function Driver Elevation of Privilege – CVE-2026-68820
Overview: A flaw in AFD.sys, the Windows kernel driver that handles low-level WinSock network socket operations, lets an attacker who already has a foothold on a machine trigger a race condition and win SYSTEM privileges.
Severity and score: High, CVSS 7.0.
Type: Use-after-free, CWE-416.
Disclosure: Check Point Research reported the flaw to Microsoft in late July 2026. Microsoft patched it as part of the August 11 Patch Tuesday release, alongside roughly 400 other fixes and two other publicly disclosed (but not yet exploited) zero-days affecting the Windows User Profile Service and the Container Isolation FS filter driver. AFD.sys was the one already under active attack, which is what makes it the entry worth reading closely.
Exploitation: Check Point traced working exploit code back to at least June 2026, deployed by Lazarus Group in a “Dream Job” style campaign against defence-sector targets. Once elevated to SYSTEM, the attackers loaded FudModule 3.1, a kernel-mode rootkit that tampers with Windows Smart App Control by zeroing out its VerifiedAndReputablePolicyState flag, then hides its own tooling from whatever endpoint security product happens to be watching.
Mitigation: Apply the August cumulative update without delay if this hasn’t already happened. Check Point’s threat intelligence director, Sergey Shykevich, put the practical advice plainly: patch the moment updates land, verify software through official channels rather than search rankings, and extend zero-trust thinking to sources that look legitimate. Lazarus’s initial access here still runs on convincing job-offer lures aimed at individuals, not a technical flaw in the delivery.
Cisco Secure Firewall ASA/FTD Remote Access SSL VPN Denial of Service – CVE-2026-20349
Overview: Cisco ASA and Secure Firewall Threat Defence (FTD) software fail to properly check for errors while processing HTTP requests on Remote Access SSL VPN interfaces. A remote attacker with no credentials at all can send crafted requests that crash the device outright.
Severity and score: High, CVSS 8.6.
Type: Improper handling of exceptional conditions during HTTP request processing (CWE-20 family).
Disclosure: Cisco published the advisory on August 11, 2026, and confirmed exploitation was already under way at the time of publication. Affected releases span ASA 9.16 through 9.24 and FTD 7.0 through 7.7, plus 10.0, a wide enough range that most currently supported deployments needed to check their version.
Exploitation: CISA added the CVE to its KEV catalog the same week, with a remediation deadline of August 14, 2026 for federal civilian agencies under Binding Operational Directive 26-04, a three-day window that leaves almost no room for change control on production VPN gateways.
Mitigation: Cisco is explicit that there is no workaround that fully addresses the flaw. Hotfixes are available per branch (ASA 9.24 needs version 9.24.1.221, for example), and installing them or upgrading outright is the only complete fix. Because there’s no clean detection signature short of the crash itself, watch for unexpected device reloads and abnormal HTTP traffic spikes hitting VPN-facing interfaces, and treat any internet-exposed ASA or FTD appliance as this month’s priority patch.
N-able N-central Authentication Bypass, Twice – CVE-2026-18556 and CVE-2026-18577
Overview: N-central, N-able’s remote monitoring and management platform used by MSPs to administer client environments, shipped with a flaw that let an attacker bypass authentication and take over an administrative account with no credentials. What makes this entry unusual is that N-able’s first attempt to fix it didn’t close the hole.
Severity and score: High, CVSS 8.2 (CVSS 4.0 scale) for both CVEs.
Type: Authentication bypass via an alternate path or channel, CWE-288.
Disclosure: N-able patched the original flaw, CVE-2026-18556, in version 2026.2. On 31 July, the company began investigating unusual licensing errors reported on customer instances still running 2026.1 and earlier, and by 2 August had confirmed an alternative exploitation path that went around its own fix entirely. A second patch, version 2026.3.1.7, addressing the newly assigned CVE-2026-18577, shipped that same day.
Exploitation: In the window between the two patches, attackers used the bypass to gain remote administrative access to N-central servers, then used the platform’s own Take Control feature, the tool built to let technicians reach client endpoints, to pivot outward and plant persistent Cloudflare tunnels on managed devices for long-term access. Finland’s national cyber security centre issued its own advisory the day the second patch landed.
Mitigation: Every N-central customer needs to confirm they’re running 2026.3.1.7 or later. The version that only fixed the first CVE isn’t enough on its own. Beyond patching, hunt for and remove any unauthorised tunnel software on managed endpoints. An RMM platform compromise never stays contained to the RMM server itself; it’s a foothold into every environment that server manages.
Progress Kemp LoadMaster Unauthenticated Command Injection – CVE-2026-8037
Overview: A flaw in LoadMaster’s escape_quotes() function mishandles input in a way that lets an unauthenticated attacker run arbitrary commands directly on the load balancer.
Severity and score: Critical, CVSS 9.6.
Type: Command injection, CWE-78.
Disclosure: watchTowr Labs published technical analysis of the flaw back in June 2026, and a patch has been available since. That didn’t stop the scanning two months later.
Exploitation: CISA logged 792 exploitation attempts against LoadMaster appliances over a 41-day window, originating from 65 unique IP addresses across 18 countries, with the most recent activity recorded on 4 August 2026. The volume was enough to earn the CVE a spot in the KEV catalog on 7 August, with a BOD 26-04 deadline of 10 August for federal systems, three days to patch a flaw that had already been public for two months.
Mitigation: Patch immediately if this hasn’t happened already. Where downtime isn’t possible right away, restrict access to the management interface, enable detailed logging, and watch specifically for command execution patterns in that log data. Given the gap between disclosure and this surge, treat any unpatched LoadMaster instance as presumed-scanned rather than merely theoretically at risk.
JetBrains TeamCity Unauthenticated Remote Code Execution – CVE-2026-63077
Overview: An unauthenticated deserialization flaw in on-premises TeamCity servers lets an attacker submit crafted data that TeamCity deserialises without properly checking its type, resulting in remote code execution with the privileges of the TeamCity server process.
Severity and score: Critical, CVSS 9.8.
Type: Deserialization of untrusted data, CWE-502.
Disclosure: JetBrains disclosed and patched the flaw in July 2026.
Exploitation: CISA added it to the KEV catalog on 5 August after confirming exploitation in the wild, with a federal remediation deadline of 8 August. Successful attacks expose stored credentials, compromise build pipelines, and hand attackers OS-level command execution on the server, a serious outcome for any organisation using TeamCity to build and ship software, since a compromised build server can quietly poison everything it produces downstream.
Mitigation: Apply the vendor patch, which adds a NoTypePermission.NONE restriction ahead of TeamCity’s allowlist so that allowlist becomes exclusive rather than advisory. Rapid7 has published a proof-of-concept tool organisations can use to validate their own detection and confirm a patch actually closed the gap, worth running given how many of this month’s other entries turned out to involve fixes that didn’t fully hold.
GitLab Unauthenticated Code Injection via GraphQL – CVE-2026-19478
Overview: A code injection flaw in GitLab’s GraphQL API lets an unauthenticated attacker modify or delete public projects and user data by sending a crafted GraphQL directive to the /api/graphql endpoint.
Severity and score: Critical, CVSS 9.4.
Type: Improper control of generation of code, CWE-94.
Disclosure: GitLab disclosed and patched the flaw on 17 August 2026, in Community Edition and Enterprise Edition versions 19.2.4, 19.1.6, 19.0.8 and 18.11.11.
Exploitation: Attackers began exploiting it around 19 August, roughly two days after disclosure, one of the fastest disclosure-to-exploitation turnarounds of the month and a reminder that “patched this week” is not the same as “safe this week” once a public advisory exists.
Mitigation: Update to a patched version without delay. Where an immediate upgrade isn’t possible, restrict unauthenticated access to the GraphQL endpoint or remove public repository access entirely in the interim, and check web logs for requests containing the string @gl_introduced, which GitLab has flagged as an indicator of attempted exploitation.
SonicWall SMA 1000 SSRF and Code Injection Chain, Weaponised by Ransomware – CVE-2026-15409 and CVE-2026-15410
Overview: Two flaws in SonicWall’s SMA 1000 series secure remote access appliances chain together into unauthenticated remote code execution. CVE-2026-15409 is a server-side request forgery in the Workplace interface that lets an unauthenticated attacker force the appliance to make network requests of the attacker’s choosing. CVE-2026-15410 is a code injection flaw in the Appliance Management Console that normally requires authentication, except the SSRF can be used to reach it without any credentials at all.
Severity and score: Critical. CVE-2026-15409 scores a maximum 10.0; CVE-2026-15410 scores 7.2 on its own, but it’s the chain that matters here.
Type: Server-side request forgery (CWE-918) chained with code injection (CWE-94).
Disclosure: SonicWall disclosed and patched both flaws on 14 July 2026, and CISA added them to the KEV catalog the same day with a three-day federal remediation deadline. On paper, this was resolved in July.
Exploitation: It wasn’t, in practice. INC ransomware became the dominant group weaponising the chain, with new victims appearing between 17 July and 1 August, and activity accelerating through the start of August. The chain lets INC affiliates pull credentials, session databases, and TOTP multi-factor seeds directly off the appliance, enough to move laterally and re-establish access even after a straightforward password reset.
Mitigation: Update to 12.4.3-03453 or 12.5.0-02835 or later immediately if this hasn’t happened. Because stolen MFA seeds and session data survive a simple patch, also rotate credentials, verify appliance integrity, and check logs for external addresses hitting /wsproxy with unusual parameters, then correlate any hits against internal authentication and lateral-movement activity.
IBM Langflow Authentication Bypass to Remote Code Execution – CVE-2026-9198
Overview: Langflow, an open-source framework for building AI and LLM workflows, ships with an /api/v1/auto_login endpoint that hands out SUPERUSER tokens without requiring any authentication. Chained with a second endpoint, /api/v1/validate/code, which executes submitted code through Python’s exec() function, the combination gives an unauthenticated attacker full remote code execution on a default Langflow deployment.
Severity and score: Critical, CVSS 9.8.
Type: Improper control of generation of code, CWE-94.
Disclosure: Langflow disclosed the flaw on 17 July 2026, covering versions 1.0.0 through 1.10.0, though independent sensors recorded exploitation attempts as early as 6 July, before the advisory even existed.
Exploitation: CISA added the CVE to its KEV catalog on 4 August. Public proof-of-concept code and a Nuclei detection template have circulated since late July, and mass scanning has continued well past that: one security firm logged 559 exploitation attempts from 310 unique IP addresses across 47 countries by early September, making this one of the most persistently scanned flaws of the summer.
Mitigation: Patch beyond version 1.10.0 immediately, and check whether any Langflow instance is reachable from the internet at all, since the default deployment is the vulnerable one. Monitor for the specific chained request pattern, a call to auto_login immediately followed by a call to validate/code, rather than relying on a single indicator to catch this.
What August Actually Showed
Eight vulnerabilities, eight different vendors, and only one of them was a genuine zero-day. That’s the pattern worth taking from this month. A criminal group weaponising a SonicWall chain that was patched back in July. A LoadMaster flaw still being actively probed two months after disclosure. A TeamCity server and a Langflow deployment, both patched for weeks, both added to the KEV catalog within the same fortnight because exploitation had finally become impossible to ignore. Even N-able’s own attempt to fix its N-central flaw needed a second try before it actually worked.
None of this argues that patching doesn’t matter. It argues that a patch shipping and an organisation applying it are two different events, and the interval between them is exactly what attackers spend the month looking for. This August’s run of KEV catalog additions reads less like a set of new discoveries and more like a lagging indicator: official confirmation of exploitation that Progress’s, JetBrains’, GitLab’s, SonicWall’s, N-able’s, and IBM’s own advisories hadn’t been enough, on their own, to drive out of production environments.
The one exception cuts the other way. Lazarus Group had working exploit code for the Windows AFD driver flaw a full two months before Microsoft’s patch existed, a reminder that state-sponsored operators don’t wait for a KEV listing to tell them what’s valuable. For everyone else this month, the vulnerability itself wasn’t the hard part. Knowing which of your already-patched systems were still exposed, and closing that gap before an August exploitation spike did it for you, was.
Where iConnect Fits In
If patch cadence alone were enough to tell you what’s exposed, none of the flaws above would still have been live targets in August. iConnect’s vulnerability assessment and managed security services scan your environment for exactly this kind of gap: systems that look patched on paper but remain reachable, still run an older build than they should, or still carry credentials an earlier compromise picked up before anyone noticed. Get in touch to find out what’s genuinely exposed in your environment before the next KEV catalog update finds it for you.
