Case file — Lead incident: A flaw in Java itself — not in any one application — lets an unauthenticated attacker run code on some of the most widely deployed enterprise servers in the world. Java4Shell (CVE-2026-114938) is a heap out-of-bounds write in OpenJDK's hardware-accelerated Base64 decoder, present in Java 17.0.6 and later, including the 21, 25, and 27 lines, on x86-64 processors with AVX-512 VBMI. Researchers at Calif demonstrated pre-authentication remote code execution against default installations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate. Oracle and the OpenJDK project shipped fixed builds on November 17, and CISA issued an alert the same day urging organizations to patch immediately. No exploitation in the wild has been confirmed, and the flaw is not in CISA's Known Exploited Vulnerabilities catalog — but the researchers have published the technique, and the patch window is now open. Status: patching under way; exploitation not yet observed.
Incident timeline
| Date | Development |
|---|---|
| November 11, 2026 | Apache releases httpd 2.4.71, fixing the critical mod_http2 race-condition flaw CVE-2026-119618. |
| November 12, 2026 | Red Hat publishes bulletin RHSB-2026-017 on the Apache flaw; public proof-of-concept repositories appear on GitHub. |
| November 16, 2026 | Infoblox confirms a breach through a compromised SharePoint migration vendor; the n0n extortion group lists 59 organizations, including Visa and JPMorgan Chase, on its leak site. |
| November 17, 2026 | CISA issues an alert urging immediate patching of Java4Shell; Oracle and the OpenJDK project ship fixed Java builds the same day. |
| November 17, 2026 | Singapore's Cyber Security Agency warns that the Apache mod_http2 flaw is being actively exploited and a public proof of concept is available. |
| November 17, 2026 | Visa confirms an incident at a third-party provider in Asia Pacific; JPMorgan Chase says its service outage was caused by a software bug, with no evidence of a cyberattack. |
| November 17, 20:35 UTC | Malicious agent-framework 1.26.1 is published to PyPI; version 1.26.2 follows at 21:08 UTC; both are removed by about 21:30 UTC. |
| November 18, 2026 | The Hacker News reports the Java4Shell details; no exploitation in the wild has been confirmed and the flaw is not in CISA's KEV catalog. |
1. Java4Shell: pre-auth RCE in the Java runtime itself hits Jira, Jenkins and Keycloak
The most consequential vulnerability disclosed this week sits below the applications it endangers. Java4Shell (CVE-2026-114938) is a heap out-of-bounds write in the HotSpot JVM's AVX-512 VBMI Base64 decoder: a malformed string ending with an extra "=" makes the decoder write 62 or 63 bytes past the end of its output buffer. Because Base64 decoding is something login forms, token parsing, and file uploads all do as a matter of routine, the flaw is reachable from the network without credentials. Researchers at Calif, who disclosed the flaw this week, demonstrated unauthenticated remote code execution against default installations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate, and authenticated code execution against Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Spinnaker.
The flaw affects OpenJDK 17.0.6 and later, including the Java 21, 25, and 27 lines, plus Oracle's Java SE 8 Enterprise Performance Pack — but only on x86-64 Intel and AMD chips that support the AVX-512 VBMI instruction set, such as Intel Xeon from Ice Lake onward and AMD EPYC from Zen 4 onward. ARM64 servers are not affected. Oracle scores the flaw 8.1 out of 10 under CVSS 3.1, rating it hard to exploit but requiring no login; Calif rates it 9.2 under CVSS 4.0. The OpenJDK project and Oracle released fixed builds on November 17 — 17.0.21.1, 21.0.13.1, 25.0.5.1, and 27.0.1.1 — and Amazon has shipped fixed Corretto builds. Other distributions, including Eclipse Temurin, Azul Zulu, and the Linux distributions, had not yet published fixes as of the morning of November 18. CISA issued an alert on November 17 urging organizations to patch immediately, starting with internet-facing identity, CI/CD, and collaboration systems.
No exploitation in the wild has been confirmed: Canada's Cyber Centre said this week it was not aware of any attacks, and the flaw was not in CISA's KEV catalog when The Hacker News checked on November 18. Calif is holding back its product-specific exploit chains and has published only the general technique — but other labs have already reproduced crashes and a memory leak. Where patching has to wait, a single JVM flag closes the exposure: -XX:-UseBASE64Intrinsics.
Investigation notes
Official anchor: CISA's alert of November 17, 2026 ("Critical Java Vulnerability Requires Immediate Patching"), plus the OpenJDK Vulnerability Group advisory and Oracle's November 2026 Critical Patch Update listing the fixed builds.
What is confirmed: the flaw, the affected versions and CPU requirement, the pre-authentication RCE demonstrations on six products (Calif), the fixed builds and their release date (Oracle/OpenJDK), and the CISA alert. The two CVSS scores differ because Oracle scores under CVSS 3.1 and Calif under CVSS 4.0 — both are reported, not reconciled.
What is not known: whether any attacker has exploited the flaw — none has been observed, and the KEV listing is absent. The researchers' product-specific chains are unpublished, so the time between patch availability and weaponization is unknowable; treat the absence of observed exploitation as a window, not a verdict.
What defenders should do: inventory every Java runtime, including the ones buried inside appliances and container images; patch internet-facing Keycloak, Jira, Confluence, Jenkins, Artifactory, and PingFederate first; where a fixed build is not yet available, start the JVM with -XX:-UseBASE64Intrinsics; and treat unexplained JVM crashes or restarts on internet-facing Java servers as possible attack attempts.
2. Apache mod_http2 flaw now actively exploited, Singapore's CERT warns
A week after Apache patched it, the critical race condition in its HTTP/2 module is being exploited in the wild. Singapore's Cyber Security Agency published alert AL-2026-164 on November 17, saying CVE-2026-119618 is "reportedly being actively exploited, and a proof-of-concept exploit is publicly available," and urging administrators to patch immediately. The flaw is a race condition in the assign_headers() function of mod_http2: under a specific request shape — body-less, with a Content-Type header and no Content-Length, when the request header table is full — a worker thread writes into memory owned by the connection thread, corrupting the heap and opening the way to unauthenticated remote code execution with the web server process's privileges. Apache HTTP Server 2.4.55 through 2.4.70 is affected, along with standalone mod_h2 2.0.8 through 2.0.42, but only when HTTP/2 is enabled. The fix shipped in httpd 2.4.71 on November 11. Red Hat rates the flaw 9.8 under CVSS 3.1; the CVE record carries 8.1.
The exploitation report should be read with its limits in mind: Singapore's CSA has not said who is attacking, where the attacks were seen, or how many servers were hit, and CISA had not added the CVE to its KEV catalog as of November 18. But the exposure is vast. Wiz reported that 64% of the cloud environments it examined had resources running affected versions, and 51% had publicly exposed instances. Shodan currently shows about 10.2 million exposed Apache httpd servers worldwide. Public proof-of-concept repositories have been available since November 12, and Red Hat says public research has shown a reliable exploit against a default-configured server with mod_http2 enabled. If patching cannot happen today, the workaround is to remove h2 and h2c from the Protocols directive and serve HTTP/1.1 only.
Investigation notes
Official anchor: Singapore CSA alert AL-2026-164 of November 17, 2026, and the Apache HTTP Server 2.4 security advisory for the 2.4.71 release. The flaw was reported by Anthropic's OSS Scanner on October 26 and independently by Google Cloud researchers Anthony Weems and Simon Scannell on November 3.
What is confirmed: the vulnerability and its mechanics, the affected and fixed versions, the public proof of concept, and the CSA's report of active exploitation. The CSA's report is credible but, as of November 18, not yet corroborated by CISA's KEV listing or by any named victim.
What is not known: the identity of the attackers, the scale of exploitation, and whether any specific organization has been breached through the flaw.
What defenders should do: upgrade to httpd 2.4.71 or a patched distribution package now; where that is not possible today, set Protocols http/1.1 in every configuration file and virtual host block; rebuild container images built before November 11.
3. Microsoft's agent-framework PyPI package compromised to steal Azure and GitHub credentials
Two malicious releases of Microsoft's official agent-framework Python package were published to PyPI on the evening of November 17, carrying a credential stealer aimed squarely at the machines that hold cloud secrets. Version 1.26.1 appeared at 20:35 UTC; version 1.26.2 followed at 21:08 UTC. Both were removed by about 21:30 UTC, and 1.26.0 is again the latest release — but removal from PyPI does not remove copies already installed or cached, and any environment that installed either version during the roughly 45-minute window should be treated as compromised.
The stealer needs no import to run. Each malicious version drops a .pth file into the site-packages root — agent_framework_meta.pth in 1.26.1, af_runtime_diag.pth in 1.26.2 — and Python executes .pth files on every interpreter start, so the collector runs whenever Python starts in that environment, even if nothing imports the package. It sweeps environment variables matching a long keyword list (TOKEN, SECRET, KEY, AZURE, AWS, GITHUB, OPENAI, HF_, FOUNDRY, PIP, NPM, NUGET, and more), reads credential files such as Azure CLI data, .npmrc, .netrc, .git-credentials, Docker configuration, and .env files, and requests fresh Azure DevOps OIDC tokens, Azure managed-identity tokens, and — in 1.26.2 — GitHub Actions OIDC tokens for the two audiences Azure federated credentials accept. Everything is sent to a webhook.site address. Wiz, which traced the uploads, said the destination used the service's free tier and that no more than 50 requests were captured — a small number, but the window was short and the package sees about 41,000 downloads a week.
The releases came through Microsoft's legitimate PyPI publisher account, not a lookalike name: Wiz found no matching 1.26.1 tag or release in Microsoft's GitHub repository and no malicious code on its main branch, which points to the publishing account rather than the source code. How the attacker obtained publishing access is not yet known, and PyPI has quarantined the publisher — though the attacker is believed to still hold publishing access, since 1.26.2 was published after 1.26.1 was pulled. This is the second Microsoft package on PyPI to receive rogue releases in six months. Anyone who installed 1.26.1 or 1.26.2 should pin agent-framework==1.26.0, rebuild affected environments, and rotate every secret the machine could reach, including Azure service principal credentials, GitHub and npm tokens, and anything in .env files.
Investigation notes
What is confirmed: the two malicious versions, their publication and removal times, the .pth persistence mechanism, and the exfiltration destination (Wiz, Aikido, SafeDep); the legitimate publisher account as the source of the releases; and the 1.26.0 safe version. The 50-request figure is Wiz's observation of the free-tier webhook endpoint, not a count of compromised machines.
What is not known: how the attacker gained publishing access, how many installs of the two versions occurred, and whether any stolen credentials have been used. PyPI's quarantine does not by itself prove the access has been revoked.
What defenders should do: check pip show agent-framework and search site-packages — including CI caches and container images built since November 17 — for agent_framework_meta.pth or af_runtime_diag.pth; treat GitHub Actions OIDC tokens as exposed on any workflow with id-token: write that installed 1.26.2; and rotate secrets before rebuilding, not after.
4. n0n names Visa and JPMorgan Chase — claims unverified, scale unclear
A newly emerged extortion group calling itself n0n listed 59 organizations on its leak site on November 16 — including Visa, Mastercard, JPMorgan Chase, Eli Lilly, Walmart, and Procter & Gamble — and set a 48-hour deadline to negotiate before publishing allegedly stolen data. The claims remain unverified: no data sample has been published, and the listings describe marketing and analytics records — cardmember names, email addresses, phone numbers, and travel and dining bookings — rather than card numbers, account logins, or balances. The scale of any compromise is unclear.
Visa has confirmed an incident, but not a breach of its own systems. A spokesperson told Cybernews on November 17 that the incident began at a third-party service provider in the Asia Pacific region: "Visa's systems were not involved and were not impacted. There is no indication that payment data such as card numbers have been affected." Visa has not named the provider or said how many people may be affected. JPMorgan Chase told Cybernews that its service disruption on November 16 came from a software bug and that it has no evidence tying the disruption to a cyberattack; it has not confirmed any connection to n0n's listing. No other listed company has confirmed a breach. n0n appeared in September 2026 and, according to SOCRadar analysis cited by The Register, has claimed about 108 victims; SOCRadar considers the group likely Russia-based, though that attribution is not definitive. The group steals data and demands payment not to leak it, without encrypting systems the way classic ransomware does — and its listings have sometimes tracked real supplier incidents: Expedia confirmed a supplier breach on November 12, while n0n claims about 830 million rows across 35 loyalty programs, a figure that counts database rows, not people affected.
Investigation notes
What is confirmed: the 59 listings on November 16 and the 48-hour deadline (Ransomtracker); Visa's third-party provider incident in Asia Pacific and its statement that its own systems and payment data were not affected (Visa spokesperson to Cybernews, November 17); JPMorgan's software-bug explanation and its statement that it has no evidence of a cyberattack.
What is not known: whether n0n holds any of the data it claims, how much data was taken from any victim, which third-party provider was involved in Visa's case, and whether the JPMorgan listing has any basis at all. A name on a leak site establishes that criminals made a claim; it does not establish access to a payment network, a bank database, or customer money.
What to watch: the 48-hour deadline expires on November 18. If n0n publishes data, the contents — not the listing — will establish what was actually taken, and from whom.
5. Infoblox breach: customer network configuration snapshots stolen via SharePoint migration vendor
Infoblox, whose DNS, DHCP, and IP address management systems run inside banks, telecoms, and government networks, confirmed on November 16 that attackers stole internal files through a compromised data migration vendor. The files include some customers' NIOS configuration snapshots — the detailed map of their DNS and network setup, including Grid databases and reporting logs — plus employees' financial and identity data. The vendor, which Infoblox has not named, was reportedly doing SharePoint migration work, and vendor accounts with trusted access were used to pull files from Infoblox's Microsoft SharePoint. Infoblox says its products, cloud services, software updates, and build-and-release processes were not involved, and it has not said how many customers are affected.
This is the third vendor-route breach covered in this brief's recent run, after the Optimizely supply-chain attack on 178 brands and the Fireworks AI breach that exposed Uber drivers' license data. The pattern is the same in all three: the attackers never touched the victim's own infrastructure — they walked in through a supplier's trusted access. Organizations running Infoblox should ask their account team whether their snapshots were taken and rotate the secrets in their Grid as a precaution.
Investigation notes
What is confirmed: the unauthorized access and the vendor route (Infoblox spokesperson, November 16); the NIOS configuration snapshots and employee financial and identity data among the stolen files; and the statement that Infoblox products, cloud services, updates, and the build process were not involved (TechCrunch, The Register, Security Affairs).
What is not known: the identity of the migration vendor, how many customers' snapshots were taken, and whether the stolen configuration data has been used. Infoblox has not disclosed a timeline for the intrusion itself.
What defenders should do: confirm with Infoblox whether your organization's snapshots were among those taken, rotate Grid secrets as a precaution, and review which third-party vendors hold trusted access to your SharePoint and document stores.
Sources
- Java4Shell Flaw (CVE-2026-114938) Enables Pre-Auth RCE in Jira, Jenkins, Keycloak and More — The Hacker News
- Critical Java Vulnerability Requires Immediate Patching — CISA
- Critical Apache HTTP Server RCE Vulnerability Actively Exploited in Attacks — Cyber Security News
- Microsoft agent-framework Compromised: Credential Stealer Targets Azure and CI Pipelines — Wiz
- Visa confirms cyberattack as hackers claim JPMorgan was hit too — Cybernews
- Infoblox Breach: Hackers Used a SharePoint Migration Vendor to Steal Customer NIOS Configuration Snapshots — ifynx
