The Exploit Bulletin

Archives
Log in
Subscribe
November 18, 2026

The Exploit Bulletin — Wednesday, November 18, 2026: 6 issues require action

Wednesday, November 18, 2026 — 6 issues require action. If you run none of the software below, you are done.

Affects: Oracle Java SE / Oracle GraalVM for JDK / Oracle GraalVM Enterprise Edition / OpenJDK · ConnectWise ScreenConnect · Perforce Puppet Enterprise · pypi/agent-framework · Starlette / Red Hat AI Inference Server · Apache Roller


1. Memory corruption in OpenJDK AVX-512 Base64 decoder enables unauthenticated RCE in Java applications (CVE-2026-114938)

REMOTE CODE EXECUTION · CRITICAL · CVSS 8.1

Oracle shipped the fix in its 2026-11-17 Critical Security Patch Update while GreyNoise is already seeing scanning and exploitation attempts against this pre-auth bug, so any unpatched Java 17+ service reachable from the internet on AVX-512 hardware is now a live target even though no successful compromise has been confirmed yet.

A memory-corruption bug in OpenJDK's hardware-accelerated Base64 decoder, present in OpenJDK 17.0.6 and later on x86-64 CPUs with AVX-512 VBMI, lets a remote attacker send malformed Base64 input to any application that decodes it before authentication and gain arbitrary code execution or crash the JVM. Researchers report pre-auth reachability in Jenkins, Jira, Confluence and Keycloak among others, and Oracle has released fixes in its November 2026 Critical Security Patch Update.

Affected: Oracle Java SE 8u511b33-perf; Oracle Java SE 17.0.21.0.3; Oracle Java SE 21.0.13.0.3; Oracle Java SE 25.0.5.0.3; Oracle Java SE 27.0.1.0.3; Oracle GraalVM for JDK 17.0.21.0.3; Oracle GraalVM for JDK 21.0.13.0.3; Oracle GraalVM Enterprise Edition 21.3.20.0.3; OpenJDK 17.0.6 and later on x86-64 systems with AVX-512 VBMI support

How to Test: Run java -version on every JVM host and container image: any OpenJDK or Oracle JDK from 17.0.6 onward (Oracle lists 8u511, 17.0.21, 21.0.13, 25.0.5, 27.0.1 and the GraalVM 17/21 builds as affected) is in scope when it runs on x86-64 hardware that supports AVX-512 VBMI; per Red Hat, JVMs on CPUs without AVX-512 VBMI do not reach the vulnerable code path, so check CPU flags on each host to prioritize. Inventory internet-facing Java applications first — Jenkins, Jira, Confluence and Keycloak are named by the disclosing researchers as reachable without authentication. For signs of attack, review web/application access logs for malformed or unusually large Base64 payloads sent to pre-auth endpoints and for unexplained JVM crashes or restarts, and compare source IPs and payloads against the indicator list GreyNoise publishes on its Java4Shell dashboard (java4shell.events.greynoise.io).

How to Patch: Apply the Oracle Critical Security Patch Update released 2026-11-17 (cspunov2026) to Oracle Java SE 8, 17, 21, 25 and 27 and to GraalVM for JDK 17/21 and GraalVM Enterprise Edition 21.3, then restart every JVM. For OpenJDK distributions, install the updated JDK build from your OS or JDK vendor (Red Hat tracks its fixed packages under access.redhat.com/security/cve/cve-2026-114938). Where an upgrade must wait, Red Hat notes the flaw is only reachable on x86-64 CPUs with AVX-512 VBMI, so moving affected workloads to hosts without that instruction set removes exposure in the interim.

Evidence: GreyNoise Labs: monitoring scanning and exploitation attempts against Java4Shell · GreyNoise live Java4Shell dashboard with indicators associated with exploitation attempts · Cisco Talos: monitoring for Java4Shell exploitation, none observed in the wild yet · BleepingComputer: Java4Shell allows unauthenticated RCE on major Java applications

Full entry with sources →


2. Missing authorization in ScreenConnect client lets session participants transfer and execute files without host confirmation (CVE-2026-84869)

CISA KEV (due 2026-09-14) · REMOTE CODE EXECUTION · CRITICAL · CVSS 9.9

Huntress published on 2026-11-16 that this flaw is being used to push and run malicious files on MSP technician accounts, ConnectWise updated its advisory the same day to say attacks continue against on-premises installs that have not disabled the named extensions and updated clients, and SecurityWeek reports over 688,000 exploitation requests against more than 40 servers — a team that has only updated the server and not the technician clients is still exposed.

The ScreenConnect client before 26.6.5 lets a participant in an active remote session transfer and execute files on the other end without authorization or host confirmation. Attackers are using it to run malicious files on MSP technician machines, which then become a foothold into every customer environment the technician can reach.

Affected: ConnectWise ScreenConnect all versions prior to 26.6.5 (client/technician components; servers are not impacted by the flaw itself)

How to Test: Verify that every technician client and access agent reports version 26.6.5 or later — updating the server alone does not update clients. In on-premises installs, check the Administration area for the extensions Labtech Plus, RMM Plus, RMM+, RMM Plus Password Link and RMM+ Password Link and treat any that are enabled as an exposure. Review session activity for file transfers and executions that were not initiated or confirmed by the host, invalidate active sessions, and assess suspected compromise with separate security tooling, since updating does not remove persistence already established.

How to Patch: On-premises: upgrade ScreenConnect to 26.6.5 from the ScreenConnect download page (a valid on-premises license is required; out-of-maintenance licenses must be renewed first), then reinstall host clients and update all access agents. Cloud: no server action is needed, but host clients must be reinstalled and access agents updated. Immediately disable the extensions Labtech Plus, RMM Plus, RMM+, RMM Plus Password Link and RMM+ Password Link and do not re-enable or reinstall them pending further ConnectWise guidance; ConnectWise cannot disable them in on-premises environments on your behalf.

Evidence: CISA KEV · VulnCheck KEV · CISA SSVC: active · Vendor confirmed · Huntress 2026-11-16: CVE-2026-84869 exploited to run malicious files on MSP technician accounts

Full entry with sources →


3. Report deserialization in Puppet Enterprise gives any CA-signed agent code execution on the Puppet Server (CVE-2026-115731)

CISA KEV (due 2026-12-01) · REMOTE CODE EXECUTION · CRITICAL · CVSS 9.4

CISA added this flaw to its Known Exploited Vulnerabilities catalog on 2026-11-17 citing evidence of active exploitation, and a Puppet Server that accepts crafted reports from any CA-signed agent — or that autosigns certificate requests — hands an attacker code execution on the server that controls the entire managed fleet.

Puppet's report-processing code deserializes untrusted rich-data types from agent reports, so a client holding any CA-signed agent certificate can execute arbitrary code on the Puppet Server as the service user via a crafted PUT request. If the certificate authority autosigns requests without strong verification, an unauthenticated network attacker can obtain a certificate and exploit it without any prior access.

Affected: Perforce Puppet Enterprise 2023.8.0 through 2023.8.12; Perforce Puppet Enterprise 2025.0.0 through 2025.12.0; Perforce Puppet Enterprise 2026.0.0

How to Test: Check the Puppet Enterprise version on each primary server: 2023.8.x below 2023.8.13, 2025.x below 2025.12.1, and 2026.0.0 are affected. Check whether the certificate authority autosigns agent certificate requests; autosigning without strong verification turns this into an unauthenticated attack. Review Puppet Server logs for report PUT submissions from unexpected or newly issued agent certificates, and review the CA's list of recently signed certificates for nodes you do not recognize.

How to Patch: Upgrade to Puppet Enterprise 2023.8.13, 2025.12.1, or 2026.0.1 as applicable. Until the upgrade is applied, disable automatic certificate signing (or restrict it to strongly verified requests) and revoke any agent certificates issued to unrecognized nodes so that only trusted agents can submit reports.

Evidence: CISA KEV · VulnCheck KEV · CISA 2026-11-17: added to KEV based on evidence of active exploitation · CISA SSVC: active

Full entry with sources →


4. Microsoft agent-framework PyPI package hijacked to steal cloud and CI credentials

SUPPLY-CHAIN COMPROMISE · CRITICAL

Malicious releases published November 17 run a credential stealer on every Python start, so any environment that installed them in the last day has likely already leaked cloud and CI secrets.

Attackers with publishing access pushed agent-framework 1.26.1 and 1.26.2 to PyPI on November 17, 2026, each adding a .pth startup hook that harvests cloud and CI/CD credentials, attempts to obtain identity tokens and sends them to a webhook.site endpoint. The code runs at every interpreter start without importing the package; PyPI has quarantined both versions.

Affected: agent-framework (PyPI) 1.26.1; agent-framework (PyPI) 1.26.2

How to Test: Search lockfiles, requirements files, pip freeze output, CI caches and container images for agent-framework==1.26.1 or 1.26.2, and look in site-packages for agent_framework_meta.pth, agent_framework_meta/_telemetry.py, af_runtime_diag.pth or af_runtime_diag.py; check egress/proxy logs for connections to webhook.site from build or runtime hosts since 2026-11-17 20:35 UTC.

How to Mitigate: Uninstall the malicious versions, pin agent-framework to 1.26.0, rebuild affected environments and images from clean caches, and rotate all cloud keys, CI/CD tokens and identity/OIDC credentials present on any host that installed them.

Evidence: Vendor confirmed · The Hacker News: Attackers compromise Microsoft's Agent Framework PyPI package · StepSecurity: Microsoft agent-framework PyPI package compromised

Full entry with sources →


5. Host-header path injection in Starlette bypasses path-based authentication, chainable to LiteLLM RCE (CVE-2026-48710)

CISA KEV (due 2026-09-16) · AUTHENTICATION BYPASS · HIGH · CVSS 6.5

SANS ISC reported on 2026-11-16 that its honeypots began seeing requests over the weekend attempting to exploit this flaw chained with the LiteLLM bug CVE-2026-42271, so FastAPI/Starlette services still below 1.0.1 that rely on the reconstructed URL path for authentication are now being probed and can be bypassed, with Wiz reporting the chain yields unauthenticated code execution.

Starlette before 1.0.1 mishandles the host portion of a request, letting an attacker inject a path into the host that is prepended to the real path when the URL is reconstructed. Where authentication or routing decisions depend on that reconstructed path, an unauthenticated client can reach protected endpoints; combined with CVE-2026-42271 in LiteLLM the result is unauthenticated remote code execution.

Affected: Starlette < 1.0.1; Red Hat AI Inference Server images prior to 3.3.5 (CUDA and ROCm)

How to Test: Confirm the installed Starlette package version in every Python service (FastAPI, LiteLLM and similar) and treat anything below 1.0.1 as affected; for Red Hat AI Inference Server, verify the image is 3.3.5 or later. Review web and proxy access logs for requests to /mcp-rest/test/tools/list and for requests whose Host header contains a path component, which SANS ISC reports as the current scanning pattern for this chain.

How to Patch: Upgrade Starlette to 1.0.1 or later in every affected service and redeploy; Red Hat AI Inference Server users should move to the 3.3.5 (CUDA) or 3.3.5 (ROCm) images from RHSA-2026:30089 / RHSA-2026:30088. If LiteLLM is in the stack, apply its fix for CVE-2026-42271 at the same time since the two are being chained.

Evidence: CISA KEV · VulnCheck KEV · CISA SSVC: active · Wiz honeypot telemetry: Starlette bypass chained with CVE-2026-42271 for unauthenticated RCE · SANS ISC: honeypot scans for the LiteLLM/Starlette chain starting the weekend before 2026-11-16 · VulnCheck XDB public exploit

Full entry with sources →


6. Pre-auth XML-RPC deserialization gives remote code execution in Apache Roller (CVE-2026-129448)

PUBLIC EXPLOIT · NO ATTACKS CONFIRMED · REMOTE CODE EXECUTION · CRITICAL · CVSS 9.8

Apache's advisory published 2026-11-17 states the flaw is being actively exploited and VulnCheck added it to its KEV catalog the same day; an internet-facing Roller instance left on 6.1.5 or earlier can be taken over without any credentials.

Roller's XML-RPC servlet accepts the library's non-standard vendor extension types and deserializes request bodies before the caller is authenticated, so an unauthenticated attacker can send a crafted XML-RPC request and execute arbitrary code on the server. The same flaw in 6.1.5 is tracked separately as CVE-2026-82384; 6.1.6 disables the extension types and closes the endpoint when XML-RPC is switched off.

Affected: Apache Roller 4.0.0 through 6.1.4 (CVE-2026-129448); Apache Roller 6.1.5 (same flaw, tracked as CVE-2026-82384)

How to Test: Check the installed Roller version: any build from 4.0.0 through 6.1.5 is affected regardless of whether XML-RPC is enabled, because before 6.1.6 the mapped XML-RPC endpoint still answered requests with webservices.enableXmlRpc turned off. Review web server and reverse-proxy access logs for POST requests to the XML-RPC endpoint from unexpected sources, particularly request bodies carrying non-standard (vendor extension) XML-RPC value types, and treat unexpected outbound connections or child processes spawned by the Roller JVM as signs of compromise.

How to Patch: Upgrade to Apache Roller 6.1.6 or later, which disables vendor extension types on the XML-RPC servlet and gates the endpoint on webservices.enableXmlRpc. Until the upgrade is done, block access to the XML-RPC endpoint at the reverse proxy or firewall; simply disabling XML-RPC in Roller's configuration does not close the endpoint on unpatched builds.

Evidence: VulnCheck KEV

Full entry with sources →


Read on the web · Every past edition

The Exploit Bulletin is free and daily. It publishes only what security teams must act on today — nothing else. Forward it freely.

Spot an error, or an exploit we missed? Reply here or email [email protected].

Don't miss what's next. Subscribe to The Exploit Bulletin:
← Newer The Exploit Bulletin — Thursday, November 19, 2026: 10 issues require action Older → The Exploit Bulletin — Tuesday, November 17, 2026: 3 issues require action
www.exploitbulletin.com
jbac.co
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.