What to patch now — 7 September 2026
Source: CISA Known Exploited Vulnerabilities (KEV) catalog — the most recently added entries as of today. Every vulnerability below has been confirmed by CISA as actively exploited in the wild, meaning attackers are already using it, not just researching it. If you (or a supplier) run any of this software, treat it as urgent.
1. Google Chrome, Edge and other Chromium browsers — a bug that can run an attacker's code from a malicious web page (CVE-2026-85046)
🛠️ Google Chromium V8 (Chrome, Edge, Opera and other Chromium browsers) · added 4 Sep 2026 · CISA fix-by date 18 Sep 2026
What it is: "Type confusion" means a bug in the browser's JavaScript engine (V8) causes it to misread one kind of data as another. An attacker can exploit that mix-up with a crafted web page to run their own code, potentially breaking out of the browser's usual sandbox protections — just visiting the wrong page could be enough, no download or click required.
Who's affected: Practically every business. Chrome, Microsoft Edge and other Chromium-based browsers are the default on most work computers, and this needs nothing more than loading a page.
What to do: Chrome and Edge update themselves automatically in the background, but the fix only applies after a restart. Check auto-update is actually on (IT teams sometimes disable it for stability) and close and reopen your browser today rather than leaving dozens of tabs running for days.
2. SonicWall SMA1000 remote-access appliances — two bugs together let an attacker in without a password (CVE-2026-83548, CVE-2026-83549)
🛠️ SonicWall SMA1000 Appliances · added 2 Sep 2026 · CISA fix-by date 5 Sep 2026 (2 days overdue)
What it is: SonicWall's SMA1000 devices give remote staff VPN-style access into a company network. One flaw ("server-side request forgery") tricks the appliance into making requests on an attacker's behalf, reaching internal functions it shouldn't; the second ("OS command injection") lets someone already logged in run their own commands on the device itself. Together, the first can be a way in and the second a way to fully take over.
Who's affected: Businesses (or their IT provider) using a SonicWall SMA1000 for secure remote access for staff working from home or on the road.
What to do: Apply SonicWall's fix — the deadline has passed. Ask your IT provider directly whether "SonicWall SMA1000" is part of your setup; remote-access gear is exactly the kind of device that quietly keeps running for years without anyone checking for updates.
3. Sangoma Switchvox business phone system — an unauthenticated attacker can take over the underlying database (CVE-2026-9586)
🛠️ Sangoma Switchvox · added 2 Sep 2026 · CISA fix-by date 5 Sep 2026 (2 days overdue)
What it is: "SQL injection" means an attacker can slip their own database commands into a request the software wasn't expecting, tricking it into running them against the backend database. Here a single crafted request, with no login required, can read, change or delete data — and potentially run further commands on the server itself.
Who's affected: Small and mid-sized businesses running Switchvox as their office phone/VoIP system — a self-hosted alternative to a hosted cloud phone service.
What to do: Apply Sangoma's fix — the deadline has passed. If you or your phone/IT provider runs Switchvox, treat this as urgent and confirm the system isn't reachable directly from the public internet in the meantime.
4. JFrog Artifactory — a default-configuration flaw hands an unauthenticated visitor admin access (CVE-2026-82329)
🛠️ JFrog Artifactory · added 2 Sep 2026 · CISA fix-by date 5 Sep 2026 (2 days overdue)
What it is: A different, more serious flaw than the Artifactory permissions issue in last week's briefing. "Improper authentication" here means that, out of the box, anyone with network access to Artifactory can become an administrator without ever logging in — the login check itself is what's broken.
Who's affected: Development teams using Artifactory to store and distribute their own software builds — the same tool flagged last week, now with a more severe issue.
What to do: Apply JFrog's fix — the deadline has passed. If your development team or a contractor runs Artifactory, treat this as more urgent than last week's issue and confirm the instance isn't exposed to the open internet unnecessarily.
5. Starlette (a building block behind many Python web apps) — a request-smuggling bug that can bypass login checks (CVE-2026-48710)
🛠️ Kludex Starlette · added 2 Sep 2026 · CISA fix-by date 16 Sep 2026
What it is: "Request smuggling" means the software can be tricked into disagreeing with itself about what web address was actually requested — an attacker sneaks an extra path into part of the request, and if part of the app decides who's allowed in based on that mismatched path, the attacker can slip past checks meant to keep them out. Starlette is a component used inside other tools (including the popular FastAPI framework), not something most businesses run directly.
Who's affected: Businesses with a custom-built web application or API — often commissioned from a developer or agency — that happens to be built on Python's Starlette or FastAPI underneath.
What to do: Ask whoever built or maintains your web application whether it's built on Python (Starlette or FastAPI specifically) and, if so, whether it's been updated — this is the kind of dependency that sits invisibly under custom software and gets missed in a routine update.
6. Kestra (workflow-automation software) — anyone can run their own commands with no login at all (CVE-2026-49869)
🛠️ Kestra OSS · added 2 Sep 2026 · CISA fix-by date 5 Sep 2026 (2 days overdue)
What it is: "OS command injection" means the software can be made to run operating-system commands it wasn't supposed to. Here an attacker needs no account at all — they can create and run their own automated workflow on your Kestra instance and have it execute commands on the underlying server.
Who's affected: Businesses using Kestra to automate data pipelines or internal workflows — a developer/ops tool, so most relevant if you run your own automation infrastructure rather than a mainstream SaaS.
What to do: Apply Kestra's fix — the deadline has passed. If your team runs it, patch now and check whether it's reachable from the public internet; this flaw needs no credentials, so exposure alone is the risk.
7. LiteLLM (a gateway many businesses use to connect to AI models) — a fake access token can open an authenticated session (CVE-2026-59822)
🛠️ BerriAI LiteLLM · added 2 Sep 2026 · CISA fix-by date 16 Sep 2026
What it is: "Improper authentication" here means one particular connection method (an "MCP" session — a newer way tools talk to AI models) accepts any bearer token as if it were valid. An attacker doesn't need the real credential, just something that looks like one, to get an authenticated session.
Who's affected: Businesses routing their own or a vendor's AI-tool traffic through a self-hosted LiteLLM gateway, rather than calling an AI provider directly — a newer piece of infrastructure as more companies experiment with AI.
What to do: If your team (or a contractor who set up your AI tooling) runs LiteLLM, apply the vendor's fix. This is a young but fast-growing category, so it's worth asking directly whether anything in your AI setup uses it — it may not be on anyone's usual patch checklist.
This is an awareness summary of public CISA KEV data, not professional security advice. CISA "fix-by" dates are US federal deadlines; for everyone else they're a strong urgency signal, not a legal obligation. Always confirm the affected versions and the fix against the vendor's own advisory.