Content substantially written or produced by an AI/LLM as part of the Malwlab editorial workflow.
Threat Intel
Lazarus Group's Operation Dream Job Wave 3: fake recruiter lures, two infection chains, the FudModule v3.1 kernel rootkit, and the CVE-2026-68820 AFD.sys zero-day. Includes tiered detection guidance.
self-hosting
Five agencies published a joint PLC advisory. Its reconnaissance playbook works against a homelab with a public IP just as well as a PLC rack. A first-person external audit of the Malwlab surface: enumeration checklist, before/after delta, and controls left behind to catch anyone scanning back.
Threat Intel
CVE-2026-64849 makes it possible against any MLflow tracking server that anyone can reach: a single unauthenticated HTTP request makes the server fetch the metadata endpoint for you and return the body.
self-hosting
I audited my self-hosted AI stack against CVE-2026-48710 (BadHost), the Starlette host-header auth bypass. One component ran an affected version and stayed protected by code, not by patch. The exact per-container checks, and the scan that exposed it.
cve-2026-59310
A five-day exploitation window, a reverse SSH backdoor, and a patch that closes the front door while the channel stays open. The vCenter triage.
litellm
An unauthenticated pre-auth SQL injection in the LiteLLM AI gateway, fixed in 1.83.7 and exploited in the wild 36 hours after the global advisory. The AI gateway is the perimeter device of the self-hosted AI stack — here is the triage, the forensics, and the standing posture.
homelab
A 20 GB GGUF from a trusted repo is not a download - it is code doing work. The lab note: why a clean build floor did not close the two open llama.cpp GGUF-parser flaws, and the no-egress intake/loader split that makes the file just a file again.
homelab
CISA added four self-hosted apps to KEV in three weeks: Metabase, MLflow, Zimbra, Gitea. The 72-hour operator checklist: exposure check, patch or compensate, compromise check, scoped credential rotation, documentation for the next incident.
Threat Intel
Zimbra 10.1.x below 10.1.20 with the optional zimbra-snmp package and snmp_notify enabled is directly exploitable - CERT Polska confirms an ongoing in-the-wild campaign, the CISA KEV remediation due date has passed, and the observed persistence is JSP web shells in Jetty webapps.
homelab
CVE-2026-65400 is a pre-auth root in macOS Screen Sharing's screensharingd, actively exploited and rescored 7.1 to 9.8. The patch is out, but the forgotten 5900 forward is the exposure: patch the Mac, verify nothing else can reach port 5900, and if it can, treat the box as compromised.
self-hosting
Copy Fail (CVE-2026-31431) lets any local user write 4 bytes into the page cache of any readable file, and one 732-byte script turns that into root. Container isolation never touched the page cache: rootless Podman stops the root, not the poisoned binary your other containers execute.
Threat Intel
Executive summary A pair of high-severity flaws in TrueConf Server lets an unauthenticated attacker take over the server and replace every client installer it distributes — a one-box supply-chain compromise. Both CVEs were added to CISA's Known Exploited Vulnerabilities catalog on 2026-08-20 with enforcement