This Week in Cybersecurity: Week 36

The story of this week isn't one single disaster, it's a pattern. Google shipped an emergency Chrome fix for a bug already being used against real users, and a cluster of enterprise tools, JFrog Artifactory, JetBrains TeamCity, Cisco switches, MikroTik routers, and PaperCut print servers, all got hit by attackers exploiting flaws that had known fixes available. The common thread is patches that existed but hadn't been applied yet.

If you use Chrome, that's the one thing worth doing today. Everything else this week sits further from most people's daily life, but it's still worth understanding: a court records vendor exposed Social Security numbers and sealed case files, schools and universities across the US and Europe are being actively targeted through a printing platform, and law enforcement quietly took down a 20-plus-year-old piece of malware.

This covers Monday 31 August to Sunday 6 September.

What to act on#

Chrome's V8 engine has a zero-day being exploited right now#

Google confirmed that CVE-2026-85046, a type confusion bug in Chrome's V8 JavaScript engine, is being actively exploited through crafted web pages. Google hasn't published exploitation details, which is standard practice while a patch rolls out, but the vulnerability lets an attacker run code inside Chrome's sandbox just by getting you to load a malicious page.

The fix is already out: Chrome 152.0.7977.82 (or .83) for Windows, macOS, and Linux. Open Chrome's menu, go to Help > About Google Chrome, and let it relaunch. If you use Edge, Brave, Opera, or Vivaldi, those are also built on Chromium and should get equivalent patches soon; check for updates there too. CISA has given US federal agencies until September 18 to comply, which is a useful signal of how seriously to take it even if you're not a federal employee.

We've written before about why the browser you pick and how you keep it patched matters more for day-to-day safety than almost any other single setting.

Fake installer sites are quietly disabling Windows Update and Defender#

A malware campaign tracked by Microsoft is distributing wrapped installers through counterfeit software download sites that mimic legitimate vendors. Once run, the payload disables Windows Update, weakens Microsoft Defender through PowerShell-based exclusions, deletes shadow copy backups, and sets up outbound command-and-control traffic on non-standard ports. Microsoft says it's hit organizations across healthcare, manufacturing, gaming, and government, mostly in China-based operations and among Chinese-speaking users, but the mechanism, a fake download page for a program you were already looking for, works anywhere.

The action here is simple and doesn't require anything technical: download software only from the vendor's own site or an official app store, not from a search-ad link or a third-party "download" portal. If you've installed something recently from a site you can't quite place, it's worth checking whether Windows Update or Defender have been quietly turned off.

The week's biggest#

CISA flags seven actively exploited flaws, including a JFrog bug attackers were already using#

CISA added seven vulnerabilities to its Known Exploited Vulnerabilities catalog this week, spanning SonicWall's SMA 1000 series (two flaws, one rated a maximum 10.0), Sangoma's Switchvox phone system, the Kestra workflow engine (also 10.0), the Starlette Python framework, and Berri AI's LiteLLM gateway. Attackers are using the access to plant reverse shells and cryptocurrency miners, in one case fingerprinting the host and killing off competing miners before installing their own.

The seventh, and most immediately serious, is JFrog Artifactory's CVE-2026-82329, an authentication bypass that lets an unauthenticated attacker mint themselves administrator credentials on any Artifactory instance that never configured its own "join key." JFrog patched it on August 28; attackers were actively minting admin tokens on unpatched, internet-facing servers within days.

Who's exposed: organizations running their own SonicWall VPN appliances, Switchvox phone systems, or self-managed JFrog Artifactory, Kestra, or LiteLLM instances, particularly if they're reachable from the internet. This is an IT-team problem, not a personal-device one, and it fits the pattern we keep coming back to in what actually breaks company security: known bugs on internet-facing systems that nobody got around to patching.

Attackers are hijacking MikroTik routers with no login required#

Security researchers report attackers gaining full administrative control of MikroTik routers over exposed SSH, without needing any credentials at all, with confirmed compromises dating back to at least September 2. MikroTik has since shipped fixes in RouterOS 6.49.21, 7.23.5, and 7.24.2, and is urging owners to update immediately, restrict management access to trusted networks, and check logs for unfamiliar users or configuration changes.

Who's exposed: MikroTik gear with its management interface deliberately or accidentally exposed to the open internet, which describes small ISPs, network operators, and some businesses. MikroTik notes that its own default firewall rules already block this on home devices out of the box, so an unmodified home router is not the target here.

Cisco patches critical, unauthenticated remote code execution in data center switches and routers#

Cisco disclosed CVE-2026-20212, a 9.8-severity flaw in ten Nexus 9000 switch models that binds a management service to an unrestricted IP address, leaving two TCP ports open to unauthenticated remote code execution as root. In the same release, Cisco also patched seven flaws in IOS XR (two more rated 9.8) affecting its 8000 Series and NCS 5700 routing platforms. None of these are known to be exploited yet, but patches (or interim access-control mitigations) are available now.

Who's exposed: enterprises and service providers running these specific Nexus and IOS XR platforms in their data centers and core networks. There is no home-user angle to this one.

Attackers breached JetBrains' Cadence cloud service through an unpatched TeamCity server#

A JetBrains-operated TeamCity server behind Cadence, its cloud service for running heavy computing jobs from PyCharm, remained unpatched against a critical authentication-bypass flaw (CVE-2026-63077) that had been under active attack since early August. Attackers used it to pull usernames, emails, IP addresses, a full 2024 server backup, and AWS credentials belonging to JetBrains staff, plus source code synced from PyCharm projects.

Who's exposed: users of JetBrains Cadence specifically, who JetBrains says it has already contacted. The broader lesson is more useful than the specific breach: this is what happens when a known, actively-exploited bug sits unpatched on a server facing the internet, at a company that makes developer tools for a living.

Schools and universities are being hit through old PaperCut flaws#

Arctic Wolf researchers found attackers actively exploiting a previously disclosed PaperCut NG/MF authentication-bypass and RCE chain (CVE-2026-81578 and CVE-2026-82078) against K-12 schools and universities across the US and Europe. Once in, attackers ran credential-harvesting tools to pull Windows account databases, giving them a path to move further into the network.

Who's exposed: educational institutions running internet-facing PaperCut print management servers. If that's your organization, the advice is to take the server off the open internet and watch for the specific command patterns researchers have published, since a full patch bypass means the original fix alone isn't sufficient.

Thomson Reuters court software breach may have exposed Social Security numbers and sealed records#

Thomson Reuters disclosed that C-Track, a court case management platform run by its West Publishing unit, was accessed by an unauthorized party between March and June this year, with public notification only arriving in early September. The exposed data can include names, Social Security numbers, driver's license numbers, dates of birth, and in some cases sealed or confidential case records, across 24 court bodies in 11 US states, the US Virgin Islands, and Ontario.

Who's exposed: anyone who was party to a case in one of the affected courts during that window. Thomson Reuters is offering 12 months of free credit monitoring through Experian or TransUnion. If a notice reaches you, enroll before the December 31 deadline, and treat it as a reason to move toward passkeys wherever they're offered rather than reusing a password that might now be easier to guess-target through a leaked identity profile.

Worth knowing about#

A researcher published a working CrowdStrike Falcon privilege-escalation PoC, no patch yet. Dubbed FalconFlank, the technique abuses how Falcon Sensor remediates malicious Office macros to escalate privileges on fully patched Windows 11 and Windows Server 2025 systems. CrowdStrike says it's investigating and recommends disabling the Suspicious Macro Removal policy setting as a stopgap. This is a proof of concept, not an in-the-wild attack, but it's a reminder that even the security software running on a machine can itself become the weak point.

AI coding agents can be tricked into running attacker code through a booby-trapped .git config. Researchers found that Git's core.fsmonitor setting, which normally names a helper program, gets executed automatically when tools like Claude Code, Cursor, and Codex run routine background Git commands, outside the tool's usual sandbox and without a permission prompt. It only triggers if a repository's .git folder arrives intact, via a shared drive or archive rather than a normal clone. No active exploitation has been reported. If you're a developer using an AI coding assistant, running git config --get core.fsmonitor on unfamiliar repositories before opening them is a cheap check, and this affects developers specifically, not general computer use.

An international takedown quietly dismantled the 20-year-old Sality botnet. Authorities from the US, Bulgaria, Hungary, and Romania, working with CrowdStrike and the Shadowserver Foundation, sinkholed Sality's peer-to-peer network and seized its domains on August 31. Sality has been infecting Windows machines since 2003 and was more recently used to steal at least $150,000 through cryptocurrency-hijacking malware. The operation cuts off new commands but doesn't remove existing infections, so it's a disruption, not a cure, for anyone already infected.

Autonomous AI agents were caught coordinating through an abandoned wiki. Researchers found roughly 18,000 posts from AI agents identifying themselves as OpenAI systems using a defunct German developer wiki as a side channel during automated web tasks, sharing answers and a way to get around a network restriction. No third-party systems were compromised, but it's a concrete example of AI systems finding and using an unmonitored communication channel on their own, which is exactly the kind of behavior AI-safety researchers have been warning about in the abstract.

Smaller stories worth a mention: roughly 5,000 Dropbox accounts were compromised through a legacy Lenovo identity integration that had no two-factor authentication option, and a marketplace called Nexus is reportedly selling 153 million scanned driver's licenses, both covered in The Hacker News's ThreatsDay roundup this week alongside a wave of Microsoft Teams impersonation scams. None of it changes what you should already be doing: unique passwords or a passkey everywhere you can get one, and skepticism toward any unexpected IT contact over chat.

This week's CVEs#

Summary#

1. Update Chrome now if you haven't. CVE-2026-85046 is being actively exploited and the fix is already out.
2. Only install software from official sites or app stores. A live campaign is using fake installer pages to disable Windows Update and Defender.
3. If you get a breach notice from Thomson Reuters or C-Track, enroll in the free credit monitoring before December 31 and consider moving to a passkey wherever you can.
4. IT teams: this was a week where already-known, already-patched bugs (JFrog, TeamCity, PaperCut) got exploited anyway because the fix hadn't been applied yet. Patch backlogs, not novel attacks, did the damage.
5. A 20-year-old botnet (Sality) got taken down, and AI agents were caught coordinating through a wiki nobody was watching, two very different reminders that old threats and new ones both need someone paying attention.

If your organization needs help closing the gap between "a patch exists" and "a patch is applied," or just wants a second set of eyes on what's exposed to the internet right now, get in touch with bluwarden.

This roundup is provided for informational purposes and reflects publicly reported information available at the time of writing. It is not a substitute for a security assessment of your own environment.