BTH News: November 19, 2026 | The Patch Window Is Shrinking
November 19, 2026
It was quite a week for technology. Microsoft 365, iCloud, Chase and even the FAA experienced disruptions. For a while, it seemed like the Internet had decided to take a personal day.
There's no evidence these outages were part of a coordinated attack. They did remind us how much we depend on systems we don't control. Reuters on the outages
This week in Between The Hacks, trusted websites delivered malicious code, an AI vendor breach exposed Uber driver information, and new NGINX and Java flaws sent teams looking for patches. A compromised firewall also helped attackers reach a VMware host.
Trusted websites were serving malicious code
On November 16, visitors to some well-known websites were served malicious code through Optimizely Web Experimentation, a service companies use to test changes to their sites. Report URI found the code in 188 of the 1,611 scripts it checked, associated with 178 organizations, including the BBC and Zoom.
You could visit a perfectly legitimate website and still end up running code written by a criminal. Researchers found code designed to skim payment card data and replace cryptocurrency wallet addresses. They also found code designed to deliver exploits targeting vulnerable iPhones and iPads. Those findings describe what the code could do. They don't tell us how many visitors lost data or how many devices were compromised.
Optimizely said an attacker used a flaw in its signup process to make unauthorized changes to customer projects. The company says it blocked the vulnerable path and removed the injected code. The presence of a malicious script on a page does not establish that the site's internal network was breached. Apple told Reuters that earlier updates had already fixed the vulnerabilities the code targeted. Reuters reporting
If you run a website, review which outside scripts load on login and checkout pages. Remove the ones you don't need and monitor the rest for unexpected changes. An enforced Content Security Policy can limit what those scripts can do. If your site was affected, follow Optimizely's customer guidance.
Optimizely incident report · Report URI investigation
An AI vendor breach reached Uber drivers
According to TechCrunch's reporting on Uber's Oregon breach filing, Uber notified 149,346 people after someone used compromised credentials from a service provider to access Uber data. The information included names, addresses, dates of birth, and driver's license numbers. An Uber spokesperson identified the provider as Fireworks AI.
Uber said its alerts didn't trigger because the intruder used the provider's own credentials. The credentials worked. The person using them wasn't authorized.
Fireworks' November 17 update distinguishes two security incidents. Cursor separately reported that an attacker obtained a key Cursor had issued to Fireworks. Cursor also found a malicious extension in its marketplace. The Uber and Cursor disclosures illustrate different downstream consequences of vendor access. The public disclosures don't establish a single continuous attack connecting them.
Earlier this week, I wrote about protecting your personal information in AI assistants. Businesses face a related question: what data and credentials can an AI supplier reach, and would you notice if someone used that access in an unexpected way?
If you receive a breach notice, verify it through a channel you already trust before following its instructions.
Fireworks AI security bulletins · Cursor's incident update · TechCrunch's reporting on the breach notices
nginx2root: check what runs in front of your website
NGINX is one of those pieces of software you may use every day without knowing it. It helps websites handle incoming traffic, and other products often bundle it.
On November 18, F5 disclosed five vulnerabilities known as nginx2root. Which flaws affect you depends on how NGINX is configured. Successful exploitation could let attackers run code with the same privileges as the NGINX worker process. CERT-EU reported a successful research demonstration for one flaw and no reported exploitation in the wild as of November 18.
If you manage web infrastructure, check your internet-facing NGINX systems first, including copies bundled with other products. The fixed upstream open-source releases are 1.30.7 and 1.31.9. For other builds, follow the vendor's update guidance.
F5 overview of the NGINX vulnerabilities · CERT-EU advisory
Java4Shell: which applications actually need attention?
Java4Shell (CVE-2026-114938) could let an attacker run code on certain servers. Exploitation requires an affected Java runtime, specific processor features, and an application that sends attacker-controlled input through the vulnerable decoder. Red Hat lists the technical requirements in its advisory.
By Thursday morning, GreyNoise had seen traffic consistent with exploitation attempts. It reported no successful intrusions on its sensors. CISA urged immediate patching. Start with exposed applications that meet the conditions, including any copies of Java bundled with the application or its container.
Red Hat security bulletin · GreyNoise activity tracker · CISA patching alert
When the firewall becomes the way in
Check Point Research documented attacks against unpatched security gateways. The attackers gained access through internet-facing Mobile Access portals and used a second vulnerability to gain greater privileges on the gateway. In one intrusion described by Check Point, they moved from the gateway to a VMware ESXi host and deployed malware designed to wipe data.
If you use an affected Check Point gateway, check the vendor's advisory for the applicable hotfix and investigate whether anyone accessed the device before the fix was applied. Include the systems the gateway could reach. A patch closes a vulnerability. It can't tell you what someone did before you installed it.
Check Point's security advisory · Check Point Research investigation
The week's takeaway
A script on your website, a key you issued to a vendor, and a firewall at the edge of your network all deserve the same question: what could someone reach if they controlled it?
That's where security decisions should begin. Start with what an attacker could actually do, then work backward to reduce the risk.




