Blog
Insights, updates, and deep dives on AI agents, decentralized standards, and the future of HOL.
BREAKING: Windows zero-day lets booby-trapped files skip the download warning (CVE-2026-96951)
How to fix CVE-2026-96951: upgrade Windows to the October 2026 cumulative update (KB5127648 on Windows 11 24H2 and 25H2). It closes a Windows Shell Mark of the Web bypass that attackers are already using.
BREAKING: New no-login code execution bug in Magento and Adobe Commerce (CVE-2026-103849)
How to fix CVE-2026-103849: upgrade Adobe Commerce or Magento Open Source to the 2026-oct build for your line (APSB26-153), and keep September's VULN-39341 hotfix in place.
HOL Guard
Launch story: RAPR AI
Run Claude Code, Codex, Ollama and other AI agents side by side on your own machine, and control them from your browser, Telegram or WhatsApp.
BREAKING: Ivanti Connect Secure login bypass is being exploited, now on CISA KEV (CVE-2026-104935)
How to fix CVE-2026-104935: upgrade Ivanti Connect Secure to 25.1.1.2, 22.8R2.5 or 25.1.3.x. Attackers are using this login bypass to run commands on the VPN, and CISA gave agencies until Oct 16.
HOL Guard
BREAKING: GitLab critical patch fixes no-login code execution on self-managed servers (CVE-2026-97229)
How to fix CVE-2026-97229: upgrade GitLab to 19.4.2, 19.3.5 or 19.2.8. A crafted regex in CI/CD config could run code on a self-managed GitLab server with no login. GitLab.com is already patched.
HOL Guard
Why I Don't Let My AI Agent Apply Terraform
agentic-boundary-terraform-skills is a skill for Claude Code, Codex, Cursor and Windsurf. It lets an AI agent work on your Terraform, with deploy credentials kept out of the agent's session.
OpenMatter Network and HOL Release Proposed Standards for Verifiable AI Compliance and Agentic Security
OpenMatter Network and HOL released proposed standards for Zero-Knowledge Boundary Compliance for Autonomous Agents. Public comment is open through October 31, 2026.
HOL
Why I built GlanceFlow: knowing what Claude Code is doing, at a glance
GlanceFlow puts one calm, plain-English checklist above the Claude Code prompt: the plan, live progress, and a clear signal when Claude needs you.
Launch story: Roomcomm
Codex and other agents from different owners share one room. Every owner reads the same transcript. A room is a link: hand it to your agent, someone hands it to theirs, and they talk.
Claude Code tried to dump your Kubernetes secret
Claude Code ran kubectl get secret db-credentials -o yaml, which exports the whole Secret. HOL Guard paused it: Allow just this once or Keep blocked.
HOL Guard
Opening a LibreOffice spreadsheet can run remote Java code (CVE-2026-63277)
How to fix CVE-2026-63277: upgrade LibreOffice to 26.2.5 or 26.8.0. A saved Calc external data link could load a remote Java driver on open; five sibling file-read, file-write and SSRF bugs are fixed in the same release.
HOL Guard
BREAKING: NetScaler SAML memory overflow hits CISA KEV
How to fix CVE-2026-88779: upgrade NetScaler ADC/Gateway to 14.1-73.41 or 13.1-64.28 (SAML SP/IdP builds)
HOL Guard
Zammad session fixation to root hits CISA KEV
CVE-2026-102489 is not practically exploitable on Zammad 7.x. For CVE-2026-102490, upgrade self-hosted DEB/RPM packages to 7.2.2 or later.
HOL GuardHOL Guard research desk
Security research for the AI agent era
Threat guides and evidence dossiers on prompt injection, MCP tool poisoning, slopsquatting, and the attacks shaping how teams ship code with agents.
Prompt injection
How hidden instructions trick coding agents into unsafe work.
Read the researchEvergreen guideMCP security
Tool poisoning, overbroad permissions, and shadow MCP servers.
Read the researchNew dossierSlopsquatting and package hallucination
Slopsquatting is the supply-chain attack in which attackers register package names that AI models reliably hallucinate; when an agent writes an import for a package that never existed, the attacker’s registered code installs instead.
Read the researchNew dossierMalicious extensions and marketplace fraud
Malicious extensions and marketplace fraud place trojanized editor extensions, plugins, or CI actions inside trusted marketplaces, harvesting source code, prompts, and credentials from every developer who installs them.
Read the research