Issue no. 42 · The security zine
Security is all about context.
Practical security thinking for systems that increasingly write, deploy, and operate themselves. Fewer buzzwords. Better evidence. Occasional memes.
CVE-2026-114045 FreeRDP FreeRDP★CVE-2026-114053 vLLM vLLM★CVE-2026-114094 Dulwich Dulwich★CVE-2026-114109 Astral uv★CVE-2026-82814 Duplicator Duplicator and Duplicator Pro★CVE-2026-100134 Adobe Acrobat for Edge★CVE-2026-105453 IBM IBM Sterling Connect:Direct for Unix★CVE-2026-105949 Red Hat Cruise Control for Apache Kafka★CVE-2026-107366 IBM IBM Langflow OSS★CVE-2026-107400 IBM IBM TRIRIGA Application PlatformPatch forecast · Oct 21
Known-exploited vulnerabilities are in the wild. Patch like it is raining.
Field report · sweep 03 · 94 contacts on the board
Advisory desk
Known exploitation already skipped the prioritization meeting.
FreeRDP FreeRDP
FreeRDP before 3.32.1 has a heap-based buffer overflow in the server-side rdpsnd channel: an integer wraparound in block-alignment arithmetic bypasses validation, so a malicious RDP client can cause an unbounded heap write during server audio playback. CVSS 3.1 8.8.
vLLM vLLM
vLLM from 0.7.3 before 0.31.0 forwards a request-supplied code_revision to Transformers AutoProcessor loading, so a client can select which revision of a model processor is imported when trust_remote_code is enabled, executing arbitrary Python in the vLLM API process. CVSS 3.1 8.1 (CVSS 4.0 9.2).
Dulwich Dulwich
Dulwich before 1.2.9 does not validate DOS drive letter prefixes in Git tree path elements on Windows, so a malicious repository with a tree entry named like a drive prefix can write files outside the worktree and reach code execution. CVSS 3.1 8.8.
Live demo · try it right here
Find leaked secrets without leaking them.
Paste code, config, or logs below. 112 detectors run entirely in your browser and return redacted findings, reasoned risk scores, and rotation guidance. Nothing uploads. Nothing phones home.
Inspect source material
No upload step exists. Files are read by your browser, scanned in memory, and discarded when the tab closes.
Findings
Ready when you are.
Add source material on the left. Results are pattern matches, not proof that a credential is active.
Latest transmissions
For the work after the scanner finishes yelling.
A field guide to agentic vulnerability management (with an actual off switch)
A practical operating model for introducing agents without confusing automation, authority, and accountability.
Finding vulnerabilities is getting cheaper. The backlog has noticed.
AI compresses the cost of discovery, but the expensive parts of vulnerability management still begin after the finding exists.
The vulnerability ticket is an interface, not a filing cabinet
A useful vulnerability record should help an owner decide and act—not merely preserve scanner output.
exhibit A: the backlog