Curated cybersecurity briefings on active threats, vulnerabilities, and emerging risks
Daily Cybersecurity Briefing (19 November 2026)
Published: Loading…
At a Glance
- CISA urged immediate patching of Java4Shell, CVE-2026-114938, an OpenJDK Base64 decoder flaw enabling unauthenticated code execution in Jira, Jenkins and Keycloak.
- F5 patched two critical NGINX HTTP/2 flaws, CVE-2026-123633 and CVE-2026-123007, allowing unauthenticated remote code execution on servers with HTTP/2 enabled.
- Microsoft agent-framework PyPI versions 1.26.1 and 1.26.2 shipped a startup-hook stealer harvesting cloud and CI/CD credentials before PyPI quarantined both.
- Malicious Optimizely JavaScript served malware to visitors of 178 organisations, including BBC, CNN, Equifax, Ledger and Zoom, on 16 November.
- Perforce Puppet flaw CVE-2026-115731 and Apache Roller flaw CVE-2026-129448 were exploited in attacks, with CISA adding the Puppet bug to its KEV catalogue.
- Ukrainian police and the FBI arrested four members of a ransomware group that extorted more than $40.5 million from at least 28 companies.
Editorial Analysis
Java4Shell (CVE-2026-114938) is a memory corruption vulnerability in OpenJDK's optimised Base64 decoder. A malformed trailing = produces a negative length and writes 62 bytes beyond the end of a heap buffer, corrupting neighbouring objects before Java rejects the input. Any Base64 value decoded by an application can carry the payload, including HTTP Basic credentials, session cookies, SAML artefacts and JWTs. Exposure depends on hardware as well as software: the vulnerable assembly runs only on x86-64 processors with AVX-512 VBMI support. The same container image can therefore be vulnerable on one node and unaffected on another. Calif reports unauthenticated code execution in Jira, Confluence, Jenkins, Keycloak, Artifactory and PingFederate. Each target requires its own heap layout and timing, however, and failed attempts usually crash the JVM. Calif has published a reproducer but is withholding its exploit chains. As of 18 November, no exploitation had been reported, and the flaw was not listed in CISA's KEV catalogue, although CISA has urged immediate patching. Oracle's fixes are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. The -XX:-UseBASE64Intrinsics flag is an unofficial workaround. The CVSS 4.0 score of 9.2 contrasts with CISA's score of 8.1.
The NGINX disclosures combine severe vulnerability ratings with differing exploitation conditions. F5's release addresses ten flaws, five of them critical. Two HTTP/2 vulnerabilities, CVE-2026-123633 and CVE-2026-123007, can lead to unauthenticated code execution. CVE-2026-123007 is the older flaw. A flow-control refund that is not capped at HTTP/2's maximum window allows an unsigned check to wrap around, sending the resulting negative length into the TLS send path as an out-of-bounds copy. Wiz demonstrated code execution with an ASLR bypass on a server serving only static files. The flaw dates back to NGINX 1.9.5 in 2015. CVE-2026-123633 has narrower prerequisites, requiring NGINX to proxy requests while connection slots are running low. The other three critical vulnerabilities affect the DNS resolver, HTTP/3 and the mail proxy module, and are unrelated to HTTP/2. Disabling HTTP/2 therefore mitigates only the first two. F5 reports no exploitation, although public proof-of-concept code exists for CVE-2026-123007. Fixes are available in NGINX Open Source versions 1.30.7 and 1.31.9.
The agent-framework compromise demonstrates how quickly a stolen publishing path can be reused. Versions 1.26.1 and 1.26.2 of Microsoft's PyPI package were uploaded 33 minutes apart on 17 November. The second appeared just over two minutes after PyPI quarantined the first, with renamed files and additional GitHub Actions OIDC token requests. A .pth startup hook runs the stealer whenever Python launches, so the package does not need to be imported. It collects environment variables, credential files and the environments of other processes. It also requests fresh Azure DevOps, managed identity and GitHub tokens, which are unaffected by rotation of stored secrets. SafeDep notes that a maintainer account, vibiret, was created on the same day and can publish to agent-framework and five related projects. PyPI has not confirmed that this account uploaded the malicious versions, and the route by which access was obtained remains unknown. The package receives approximately 190,000 downloads per month, but the number of installations of the compromised versions has not yet been established.
Highlights of the Day
Java4Shell Flaw in OpenJDK Base64 Decoder Enables Remote Code Execution
A critical memory corruption bug, CVE-2026-114938, affects OpenJDK's optimised Base64 decoder on x86-64 processors with AVX-512 VBMI support. A malformed trailing "=" produces a negative result that writes 62 bytes past the buffer, letting attackers execute code in products such as Jira, Jenkins and Keycloak. Oracle has released fixes, and no exploitation in the wild had been reported as of 18 November.
Malicious Microsoft Agent Framework Releases on PyPI Steal Cloud and CI/CD Credentials
Two PyPI releases of Microsoft's agent-framework, versions 1.26.1 and 1.26.2, were published on 17 November with a credential stealer hidden in a .pth startup file. It runs whenever Python starts, without importing the package, and collects environment variables, credential files and fresh cloud identity tokens. The data is sent to a webhook.site endpoint, and PyPI has quarantined both versions.
Critical NGINX HTTP/2 Flaws Allow Unauthenticated Remote Code Execution
F5 patched CVE-2026-123633, a use-after-free, and CVE-2026-123007, a flow-control integer underflow, in NGINX's HTTP/2 handling. Both carry a CVSS 4.0 score of 9.2 and can let unauthenticated attackers execute code on servers with HTTP/2 enabled, including by bypassing ASLR. Fixes arrived in NGINX Open Source 1.30.7 and 1.31.9, and public proof-of-concept code exists for CVE-2026-123007.
Ukrainian Police Raid Ransomware Gang That Extorted Over $40 Million
Ukraine's National Police and the FBI conducted 66 searches across Kyiv and 11 regions, targeting a group of at least 27 people that includes former Conti members. The group used AI-generated voice calls posing as IT support to obtain employee credentials or install remote access software, then stole data and deployed ransomware. At least 28 companies paid over $40.5 million, and four key suspects are in custody.
NSA Internal Network Outage Continues Into Second Day
The NSA's classified internal network, NSA Net, suffered intermittent outages on Tuesday, prompting the agency to stagger staff arrivals at Fort Meade and close its campus to most employees. A source said the network is working again, but the login process is straining the system, and full restoration may take another day or two. Director of National Intelligence Jay Clayton attributed the disruption to maintenance affecting PHP and said it was not a coordinated or state-sponsored attack.
Daily Coverage