Blog
Insights, updates, and deep dives on AI agents, decentralized standards, and the future of HOL.
CVE-2026-6791: glibc wordexp Stack Clash via Tilde Expansion
CVE-2026-6791 is a stack-based buffer overflow in glibc's wordexp() function. The parse_tilde internal function uses strndupa to allocate stack memory sized by attacker input with no bounds check, enabling a stack clash.
BREAKING: CVE-2026-18165 - @fastify/oauth2 Login CSRF via Plantable State Cookies
@fastify/oauth2 7.2.0 through 8.2.0 accepts plantable OAuth state cookies from related hosts, enabling login CSRF. Upgrade to 8.3.0 and enable hostPrefixedCookies.
Michael Kantor
BREAKING: CVE-2026-18500 - @fastify/jwt Key Override Authorization Bypass
CVE-2026-18500 lets @fastify/jwt before 10.2.2 override a route-specific verification key with the global secret, breaking JWT authorization-domain separation. NVD scores it 8.1 HIGH.
HOL
CVE-2026-72842: OpenWrt luci-app-lxc ACL bypass to root code execution
CVE-2026-72842 lets a low-privileged LuCI user reach admin-only container routes in OpenWrt luci-app-lxc, then chains path traversal in the lxc_name parameter to control lxc.hook.start-host and execute code as root on the host.
Michael Kantor
CVE-2026-19135: OpenNMS JEXL Measurement Sandbox Bypass
A JEXL sandbox bypass in the OpenNMS Measurements REST API lets a low-privileged authenticated user load arbitrary Java classes on the server (CVSS 5.4, CWE-470). Upgrade to Horizon 36.0.3, Meridian 2024.3.12, or Meridian 2025.0.9.
Michael Kantor
CVE-2026-49819: UpSnap Initial-Superuser Takeover Chained to Root RCE
UpSnap 4.4.1 through 5.3.5 lets an unauthenticated network-adjacent attacker claim the initial superuser account on a fresh install, then execute shell commands as root through the wake command handler. Upgrade to 5.4.0.
Michael Kantor
Keyv Supply Chain Attack: 2 Billion Monthly Downloads Compromised in npm Worm
Attackers compromised the GitHub account of keyv maintainer jaredwray, injecting a credential-stealing worm into 434+ npm packages with 2B+ monthly downloads.
HOL Guard
Introducing plori
A maintainer's guide to the plori plugin: what it does, who it helps, and how it pairs with Guard.
CIGAR: A Governed Context Runtime for AI Agents
CIGAR Honey is an open-source runtime for compiling governed, policy-aware context for AI agents. It creates bounded context bundles with provenance, authorization, replayable evidence, scoped handoffs, and intent-first external actions.
HOL Guard 2.1: 51,000 Test Cases, HMAC-Backed Reconnect, and a New Command Classification Engine
HOL Guard 2.1 ships a 51,000-case command classification corpus, HMAC-backed dashboard reconnect, signed Codex hook manifests, DNS-pinned archive downloads, and a typed GitHub capability model — all open source, canary-tested on TestPyPI.
MCP Tool Poisoning: How the AI Agent Protocol Became a Supply Chain Attack Surface
The MCP protocol connects AI agents to over 10,000 tools. It also creates a new supply chain attack surface: poisoned tool descriptions that silently hijack agent behavior. Here's the data, the CVEs, and what to do.
HOL
Prompt Injection Defense in 2026: What Actually Works When Detection Is Impossible
OpenAI says perfect prompt injection detection is still unsolved. Three academic proofs show why it may never be possible. This is the defense-in-depth architecture that actually works in 2026.
HOL
Slopsquatting: When AI Hallucinations Become Supply Chain Attacks
AI coding assistants hallucinate package names 19.7% of the time. Attackers register those names on npm and PyPI before real packages can claim them. Tens of thousands of developers have already installed malicious packages their AI suggested. Here is how the attack works, what the research shows, and how to stop your team from becoming the next victim.
HOL