The Tools That Run the Network Behind the Network
Most people never think about Crosswork or Secure Workload. That’s kind of the point — these aren’t apps you use. They’re tools that big companies use to watch, manage, and lock down everything else. Crosswork keeps an eye on huge chunks of a company’s network. Secure Workload is a “zero-trust” system. It decides which parts of a company’s servers are allowed to talk to which other parts.
In other words, these are control-panel tools. If someone breaks into the control panel, they don’t just get one machine. They could get a map — and a set of keys — to a lot of machines at once. That’s why what Cisco found during a routine check of its own systems is worth paying attention to, even though nobody outside Cisco went looking for it.
On August 19, 2026, Cisco’s security team shared two reports covering nine flaws across these two tools. Five of them scored a perfect 10.0 out of 10 on the CVSS scale — the standard way to measure “how bad is this, really.” A 10.0 is as bad as that scale goes.

All nine flaws and their severity scores, grouped by which product they affect.
How Bad Are We Actually Talking?
A CVSS score isn’t just a scary number. It tells you specific things: how easy the flaw is to reach, whether you need a password first, and how much damage it can do once triggered. A 10.0 usually means an attacker doesn’t need to log in, doesn’t need to trick anyone into clicking anything, and can cause serious damage from far away.
The Crosswork flaws affect three tools — Crosswork Data Gateway, Crosswork Network Controller, and Crosswork Planning. Cisco says the flaws are there no matter how the device is set up. One is a SQL injection bug — an attacker could sneak database commands into a request the system wasn’t built to expect. Another lets someone reach a sensitive part of the system without logging in at all. A third lets an outsider change which files the system touches.
Over on Secure Workload, the pattern is similar but wider — Cisco says both the cloud version and the on-site version are affected. The two worst bugs there are about the system not being able to tell who’s allowed to do what. SecurityWeek added one more detail: if someone actually used these flaws, they could run their own code on the system, get past logins, dig into files they shouldn’t reach, or even delete files. That’s real, deep access — not just peeking around.
Why the Same Headline Covers So Many Bugs
Cisco’s own report explains something that might seem confusing at first: Cisco doesn’t file one report per bug. When their own testing finds several bugs that come from the same root cause — say, several different spots where SQL injection is possible — they group all of them under one ID, tied to a shared category called a CWE (Common Weakness Enumeration).
That’s genuinely useful for people trying to fix things — one report, one fix, one version to upgrade to. But it also means “nine flaws” almost certainly stands for more than nine actual bugs underneath. Cisco’s own wording calls several of these “a set of” flaws, which is the giveaway.

How Cisco bundles related internal findings into a single CVE.
The Full List, in Simple Words
| Product | CVE | CVSS | Type | What it means |
| Crosswork | CVE-2026-20030 | 10.0 | SQL injection | An attacker can sneak database commands into a request the system didn’t expect. |
| Crosswork | CVE-2026-20357 | 10.0 | Missing authentication | A key part of the system can be reached without logging in at all. |
| Crosswork | CVE-2026-20358 | 10.0 | External control of file system | An outsider can change which files the system reads, writes, or deletes. |
| Crosswork | CVE-2026-20359 | 9.9 | Weak login protection | Stored login details aren’t locked down the way they should be. |
| Secure Workload | CVE-2026-20231 | 9.9 | Command / OS / argument injection | Several ways to sneak unwanted commands into the system. |
| Secure Workload | CVE-2026-20315 | 10.0 | Weak access control | The system can’t reliably tell who’s allowed to do what. |
| Secure Workload | CVE-2026-20317 | 10.0 | Weak authentication | Covers missing logins, ways around logins, and trusting input it shouldn’t. |
| Secure Workload | CVE-2026-20318 | 9.6 | Bad input checking / path traversal | Bad input can make the system read or write files it shouldn’t touch. |
| Secure Workload | CVE-2026-20319 | 7.5 | Buffer overflow | Bad data can overflow memory the program didn’t set aside for it. |
Sources: Cisco Security Reports (cisco-sa-hardening-crosswork-UzDTU9Vh and cisco-sa-hardening-csw1-shSvndWP), The Hacker News, and SecurityWeek, Aug 19–21, 2026.
Nobody’s Been Hit — Yet
Here’s the good part: Cisco found all nine of these flaws itself, through its own testing — not because someone was caught using them. “These vulnerabilities were found during internal testing and are not known to be actively exploited,” the company said in both reports. There’s also no sign that anyone outside Cisco knew about them before the fix came out.
That matters a lot. It’s the difference between a calm, planned fix and a rushed response to a live break-in. Right now, this is the calm kind. But “not known to be used yet” doesn’t mean “safe to ignore.” Once a fix and its report are public, attackers often study the fix to figure out exactly what was broken. That’s a big reason security teams are told to patch fast instead of waiting.
There’s no quick fix or setting you can change for any of these nine flaws — the only fix is to upgrade. Crosswork Data Gateway, Network Controller, and Planning need version 7.2.1-SP or later. Crosswork Workflow Manager needs version 2.1.1-SP or later (more on that below). Secure Workload needs to move to 4.0.4.16 (for the 4.0 line) or 3.10.9.1 (for the 3.10 line and earlier). One extra thing to know if you use Secure Workload’s cloud version: Cisco already fixed its own cloud part, but that alone doesn’t finish the job. You still need to separately upgrade the Agent and Connector software running on your own end.
The Detail Buried at the Bottom of the Report
Here’s something that didn’t make the headlines but is sitting right there in Cisco’s own report, in a short part called “Source”: “These vulnerabilities were found during internal security testing using existing testing processes as well as frontier AI models.” That exact line shows up in both the Crosswork and Secure Workload reports — and in the SD-WAN/IOS XE report from two weeks earlier too.
In other words, Cisco isn’t just doing the same manual code checks it always has. Somewhere in that “internal security review,” AI models were helping hunt for bugs. Cisco has talked about this shift openly elsewhere: in a company blog post from late June 2026, Cisco said its early use of AI models showed that the old way — one report per bug — was “reaching a breaking point,” because AI-assisted searching was finding bugs faster than the old process could handle. That same post said Cisco was moving to a new schedule, sharing security fixes twice a month starting in July 2026, just to keep up.
That changes how you might read this whole story. These nine flaws probably aren’t nine bugs a person happened to notice by hand. They’re more likely the result of AI tools searching through Cisco’s own code faster and more closely than a person could alone, then getting sorted, grouped, and shared in this new, higher-volume way.
One outlet, The Register, went a step further. It pointed out that Cisco takes part in Anthropic’s Project Glasswing — a program that gives a small number of trusted groups early access to Anthropic’s strongest bug-finding AI models. The Register guessed this could be the tool Cisco used here, though it’s worth being clear: that’s The Register’s own guess, not something Cisco’s report confirms. Cisco’s own text just says “frontier AI models,” without naming who made them.
Reading the Fine Print: What CVSS 10.0 Actually Means
Both reports share the exact formula behind that 10.0 score, and it’s worth breaking down once — it shows exactly why these bugs are rated as high as the scale goes. The code reads: AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H.
- AV:N — Attack Vector: Network. The attacker doesn’t need to be near the device. This can be triggered from far away, over the network.
- AC:L — Attack Complexity: Low. No tricky timing or special setup is needed to pull it off.
- PR:N — Privileges Required: None. The attacker doesn’t need any account or access to begin with.
- UI:N — User Interaction: None. Nobody inside the company needs to click anything or fall for a trick.
- S:C — Scope: Changed. The flaw can reach beyond the one part of the system where it started.
- C:H / I:H / A:H — Confidentiality, Integrity, and Availability are all rated High. That means data can be read, data can be changed, and the system can be knocked out or slowed down.
Put together, that’s about as bad as it gets: no login needed, no one tricked, no special timing, and full damage possible once it’s triggered. That’s exactly why these bugs sit at the top of the scale.
One More Product Just Got Added to the List
There’s a small but important update hiding in the Crosswork report’s history. When Cisco first shared it on August 19, 2026, it listed three affected tools. Two days later, on August 21, Cisco quietly updated the report and added a fourth: Crosswork Workflow Manager, versions 2.1.1 and earlier, fixed in 2.1.1-SP. If your team uses Workflow Manager, that’s a fix that wasn’t on the original list — easy to miss if you only read the report once, early on.
Not a One-Off Patch Tuesday
Zoom out, and this fits a pattern that’s been building all summer. Cisco’s security team has been running what it calls a “comprehensive internal security review” across several product lines. This is at least the second big batch of results from that review — about two weeks earlier, the same process turned up 12 separate bugs across Catalyst SD-WAN and IOS XE Software, Cisco’s core software for routers and switches.
This internal review is happening at a time when Cisco gear has had a genuinely rough few weeks out in the real world, not just in testing. Earlier in August, Cisco confirmed that a different flaw in its ASA and FTD firewall software (CVE-2026-20349) was already being used by attackers. And in July, a separate flaw in Cisco’s Firepower Management Center (FMC) was also confirmed under attack before a fix even existed.

The bigger run of Cisco security news this summer — fixes found in-house, alongside separate attacks confirmed out in the wild.
Why Cisco Specifically Is Such a Popular Target
None of this happens by chance. Cisco’s own reporting on these issues points out something worth sitting with: because Cisco gear sits inside such a huge share of company networks worldwide, it’s a magnet for attackers. Flaws in Cisco products have been used again and again to break in and spread harmful software at scale. That’s not a knock on Cisco’s engineering — it’s simply the cost of being the equipment so much of the internet quietly runs on.
What This Means For You
- If your team runs Crosswork or Secure Workload, patching isn’t optional — there’s no quick fix, and five of these nine flaws are as bad as a flaw can get.
- If you use Secure Workload’s cloud version: don’t assume Cisco’s cloud-side fix covers you. Check whether your own Agent and Connector software still needs a manual upgrade.
- “Found in-house, not yet used by attackers” is a genuinely good sign, but that window tends to close fast once the report goes public. Treat this as a now problem, not a someday problem.
- This sits inside a bigger pattern of real attacks on other Cisco products this summer (ASA/FTD, FMC). If your team runs any Cisco security or network gear, now’s a good time for a broader check of what’s patched — not just these two tools.
Frequently Asked Questions
Is this actively being used by attackers right now?
No — Cisco says these nine flaws were found through its own testing and are not known to be used in the wild. That’s different from two other recent Cisco issues (the ASA/FTD flaw and the FMC zero-day), which were confirmed under active attack before those fixes came out.
Do I need to do anything if I don’t use Crosswork or Secure Workload?
Not for these nine specific flaws. But if your team uses any Cisco networking or security gear, it’s worth checking your patch status anyway — Cisco has had an unusually busy few weeks with both in-house discoveries and real attacks across its products.
Is there a quick fix if I can’t upgrade right away?
No. Cisco’s reports are clear that there’s no quick fix or workaround for any of these nine flaws. Upgrading to the fixed version is the only real solution.
Why does one report sometimes cover several different bugs?
Cisco groups related bugs it finds in-house by their root cause (called a CWE) and gives one ID per group, instead of one ID per bug. This makes it easier to share and fix things, but it also means the real number of bugs is likely higher than “nine.”
What’s the difference between the CVSS 10.0 flaws and the lower ones?
All nine are serious, but the 10.0-rated flaws (five of the nine) generally mean an attacker doesn’t need to log in and can cause severe damage from far away — the top of the scale. The lower-scored ones, like the 7.5 buffer overflow bug, are still serious but usually need more specific conditions to be used.
Is this related to the Grok or npm stories?
Not directly — different companies, different kind of technology entirely. But it fits the same bigger pattern running through security news this month: big, widely-used systems keep turning up serious flaws, whether it’s AI chat tools, open-source code, or company network gear. The common thread isn’t about any one company — it’s about how much room for error modern, connected software creates.
Did Cisco actually use AI to find these bugs?
Yes, at least partly. Both reports say these flaws were found using “existing testing processes as well as frontier AI models.” Cisco has said publicly that AI-assisted searching is happening at a scale that’s already changing how it shares fixes — it moved to sharing fixes twice a month starting in July 2026, partly for this reason.
Is this connected to Anthropic’s Project Glasswing?
Maybe, but it’s not confirmed. The Register pointed out that Cisco takes part in Project Glasswing, which gives select groups early access to Anthropic’s strongest bug-finding AI models, and guessed that could be the tool used here. Cisco’s own report doesn’t name a specific company or model, so this link is an educated guess from that outlet, not something Cisco has said directly.