Tech • AI • Robotics • Game

ENFR

If you’ve deployed anything since 2015, I have bad news...

8/10
AIFireshipNovember 20, 2026 at 07:49 PM5:35
Audio player

⏱ ≈ 3 min read · 6-min video — 3 min saved

TL;DR

A three-day stretch exposed major weaknesses in Optimizely, Java 17+, and NGINX, enabling malware delivery and remote code execution across widely used internet infrastructure.

KEY POINTS

Optimizely compromise hit 178 major sites

Attackers abused a flaw in the Optimizely signup flow to inject malicious JavaScript into other customers’ projects, which was then served from Optimizely’s own CDN. Affected sites reportedly included GitLab, Salesforce, Equifax, and Ledger, along with 174 others. Anyone visiting those pages could receive the hostile script automatically.

Malware was rapidly iterated

The payload was rebuilt 26 times in about 10 hours, showing an unusually fast operational tempo. The code was designed for skimming payment cards and stealing cryptocurrency, and it was also described as capable of additional abusive behavior. Researchers said the campaign ultimately netted only about $7,500, despite the broad reach.

Java bug enabled unauthenticated code execution

A separate disclosure detailed a remote code execution flaw in Java 17 and later tied to Base64 decoding. On newer Intel and AMD chips, the JVM can switch from standard Java code to an optimized handwritten assembly path using AVX-512. That fast path contains a bounds error that can write 62 bytes past the end of a buffer when given a short, malformed, padded input.

Common enterprise software was exposed

Because many applications decode Base64 from untrusted users, the flaw was shown to be exploitable without logging in on products such as Jira, Confluence, Jenkins, and Keycloak. The bug affects systems running Java 17+, meaning older Java deployments may avoid this specific issue. Some exploit paths reportedly became harder when NGINX was placed in front of the Java service.

NGINX flaw had existed since 2015

Another disclosure described remote code execution in NGINX, software used on more than 225 million websites. The bug sits in HTTP/2 flow control and reportedly dates back over a decade. It stems from failure to enforce the protocol’s 2 GB credit cap when cancelled requests refund unused credit back to a connection.

Integer underflow opened the path to takeover

By opening a connection, exhausting its credit, and repeatedly cancelling requests, an attacker can push the internal counter past the cap. NGINX then subtracts that value from the limit using an unsigned integer, causing an underflow that wraps into a large positive number. That allows more credit to be added until another overflow causes a negative length to be interpreted during response handling, leading to out-of-bounds memory access.

Exploit chain could lead to full server compromise

Once memory corruption is achieved, attackers can leak process memory, locate useful addresses, overwrite a function pointer triggered when a connection closes, and then disconnect to execute commands. The researchers who found the flaw also paired it with a second Linux kernel vulnerability to escalate privileges to root on the host machine. That combination turns a web-server bug into a full-system takeover.

Practical exploitation may depend on deployment setup

The public NGINX exploit reportedly needs to push around 128 GB over a single connection, which could make exploitation unreliable behind some proxies or content delivery layers. Systems fronted by services such as Cloudflare may be less exposed to that specific proof of concept. Even so, directly exposed servers and self-managed infrastructure face a more immediate risk.

CONCLUSION

The week’s disclosures showed how a marketing script platform, a core runtime, and one of the world’s most deployed web servers can each become high-impact attack paths. For organizations running public-facing services, patching and architecture review are now urgent operational priorities.

Ask a question

More from AI