Earlier this week, Kim Jong-un's wildest fantasies came true because in the span of three days, hackers turned CNN, the BBC, and Zoom into malware routers. Researchers took over Java servers with a single B 64 string, and your favorite CIS admin's favorite web server turned out to have a remote code execution bug that's been around since 2015. So, if you've deployed anything to the internet in the last decade, there's a good chance that at least one of these is now your problem. And that's not even including whatever else probably got hacked between when I recorded this audio on Thursday and when you'll see the video on Friday. In today's video, we'll break down how each attack actually works, but find out why hacking 178 major websites only paid out $7,500 and learn how to tell if your server is cooked. It is November 20th, 2026, and you're watching the code report. But things got off to a hot start on Monday when anyone visiting GitLab, Salesforce, Equifax, Ledger, or any of 174 other popular websites was fed malware. What they all had in common was that they were loading a script from Optimizely, which is an AB testing tool marketing department used to feel like they're actually contributing something to the company. A bug in Optimizely signup flow let an attacker write their own JavaScript into other customers projects, which Optimizely then happily delivered from their own CDN. The payload would skim your card, steal your crypto, text your ex, and if you were on an older iPhone, it could hit you with a zero-click exploit kit that chained together six different vulnerabilities to take over your device. Within the span of about 10 hours, the attackers rebuilt the payload 26 times, which with that level of efficiency makes you wonder if they were using Optimizely to AB test their malware. But thankfully, crypto doesn't actually have any utility. So, along with any cards they skimmed, the hackers reportedly only walked away with about 7,500 bucks. From there, the internet had about 12 hours to recover before Tuesday when a security firm called Khif disclosed a remote code execution bug in Java itself. It's called Java for shell and the bug is related to how Java decodes B 64. Usually that's handled by normal Java code. But if you're running a newer Intel or AMD server chip, the JVM gets fancy and swaps in a handwritten assembly version that uses AVX 512 to go faster. The problem is that version has a math bug where if you hand it a short string with messed up padding, it'll write 62 bytes past the end of the buffer before it figures out the input was garbage. And because Java apps decode B 64 from total strangers all day long, Khif says they got remote code execution without logging in on Jira, Confluence, Jenkins, and Keycloak. But thankfully, it only affects Java 17 and up is so if you've been practicing security through laziness, you'll be fine. Akalif did mention that some of their exploits break when EngineX is sitting in front of the Java app. And that would have been great news until Wednesday when researchers disclose that anyone with your IP can run commands on your EngineX server. If you're in the marketing department, EngineX powers more than 225 million websites, which makes it one of the most heavily audited CC code bases in the world. And yet, this bug has been around for over a decade. It lives in the part of HTTP2 called flow control, which exists so a server doesn't send data faster than the client can handle it. The way it works is the client hands out credit which is just the number of bytes Engine X is allowed to send and the protocol says that credit can never go above two gigabytes. The problem is Engine X forgot to enforce that limit when a request was cancelled. So any unused credit was refunded back to the connection without ever being checked against the 2 gig cap. So if an attacker opens a connection, maxes out its credit, and then cancels a few requests, the refunds push the counter over the cap. Now, that alone isn't a big deal, but EngineX checks the limit by subtracting the current credit from the 2 gig max. And because the counter is an unsigned integer that can't go negative, the result underflows and wraps around and becomes positive. At this point, Engine X thinks the connection is nowhere near the cap, which lets the attacker keep piling on credit until they trigger an overflow where a different part of the code reads the counter as a negative number. At that point, Engine X tries to send back a response, but the length is negative. And because the length decides how many bytes get copied, the copy runs right past the end of the buffer and starts reading and writing memory that belongs to other parts of the program. And from there, it's a fairly standard exploit where the attacker leaks enough memory to figure out where things are, overwrites a function X runs whenever a connection closes and then hangs up. So Engine X runs their command instead. The bug was found by a security firm called Zint, who then paired it with a second bug in the Linux kernel that upgrades the attacker to root on the entire machine. And while Zint was waiting on a patch, weirdly, four other teams found and reported the same EngineX bug on their own. The good news is the public exploit has to send 128 gigs over a single connection, which will probably fail behind a proxy. So, your side project on Cloudflare is likely fine. And before you ask, we're still investigating how this affects Horse Tinder servers. But while those are offline, we've been using the downtime to redesign the entire app with Paper, the sponsor of today's video. When your coding agent builds a UI by itself, it's flying blind because it can't see your screen or your team's design system. The paper is a design tool that fixes this with a canvas that is real HTML and CSS. And its MCP server lets claude code, cursor, and codecs read and write to it directly. It lets your agent pull your CSS variables in as design tokens, then draw a new screen on the canvas with your actual colors and fonts, like it's doing here to create a match screen that horse Tinder users can be proud of. If something looks off, you just grab it and fix it yourself. They then tell your agent to build it into your app. And because it's already HTML and CSS, there's no translation layer between what you approved and what actually ships. It's used in production by teams at RAMP, Lovable, Linear, and more. And you can try it out for free today at the link below. This has been the Code Report. Thanks for watching, and I will see you in the next one.
Generated algorithmically for Search Engine
Indexing.