Blog
Articles on application and infrastructure security — secure pipelines, authentication, and practical security engineering.
65 articles
Anthropic's own dashboard shows 6,157 vulnerabilities disclosed and 516 patched. Why the bottleneck is human, where the numbers are contested, and what to do.
Read More
The Cyber Resilience Act reporting duty has applied since 11 September 2026. Here is the 24h, 72h and final-report clock, ENISA's platform and open-source dependencies as a runbook.
Read More
What NIST's CAISI AI Agent Standards Initiative covers: three pillars, the security RFI, the NCCoE agent identity paper, and what engineers can do now.
Read More
A 97,940-model study shows generated AI bills of materials are structurally complete but documentation-poor. The numbers, the limits of the method, and what to add by hand.
Read More
A clean Hugging Face pickle scan is not proof of safety. What the ShadowPickle paper measured across ten scanners and four hubs, and a safetensors-first loading policy you can enforce.
Read More
Indonesia's PDP implementing regulation read against the 2022 law: which of the 3x24-hour notice, DPO and 2% fine rules are new, and what to build before January 2027.
Read More
How ClawHavoc spread 341 malicious SKILL.md skills on ClawHub with fake Prerequisites, and a vetting script plus Claude Code settings to defend agents.
Read More
GitGuardian counted about 29 million new public secrets in 2025 and 24,008 inside MCP configs. What the numbers prove, and a plan to scan developer endpoints.
Read More
What Anthropic's September 2026 misuse report says about agent-driven intrusions, stolen session tokens and AI API keys, and the controls a DevOps team can check this week.
Read More
Veracode's 2026 report says AI-generated code passes security tests 56% of the time. Which models and CWE classes fail, the data limits, and how to gate it
Read More
CVE-2026-22708: Cursor ran export and typeset without approval, so a poisoned PAGER made an allowlisted git command run code. Fix 2.3, scores and defender steps.
Read More
Mandiant says a hijacked AI coding assistant session led to a poisoned PyPI package, stolen GitHub tokens and a worm in about 100 repos. What is known, and the defences.
Read More
The keyv npm worm shipped through GitHub Actions and OIDC with valid provenance. What provenance proves, what it cannot, and which gates to add instead.
Read More
Regulation (EU) 2026/1744 moved AI Act high-risk dates to Dec 2027 and Aug 2028, but Article 50 and two Dec 2026 deadlines still apply. A DevOps calendar.
Read More
A Shai-Hulud variant hit the Tensorlake npm SDK through a preinstall hook and the project's own release workflow. What it took, and how to block it.
Read More
A May 2026 re-test of five frontier models found package hallucination rates of 4.62% to 6.10%, yet 127 names were invented by all five. What to gate before install.
Read More
A close reading of the Deadbugz MCP campaign: a server that rewrote its own tool metadata after the third call, and how to fingerprint definitions and gate changes.
Read More
SVD-2026-0808 read closely: why a 9.1 MCP Server flaw that needs admin rights still matters, the AI Toolkit CVEs, and a version-gate routine for patch triage.
Read More
How GitLost used a public issue to leak private repos via a GitHub Agentic Workflow, and the allowed-repos, min-integrity and safe-outputs limits
Read More
How one GitHub issue reached CI secrets in Claude Code Action, Gemini CLI and Codex workflows, with patched versions and a hardening checklist.
Read More
How a crafted image defeats runAsNonRoot in containerd, which versions fix it across 1.7, 2.0, 2.2 and 2.3, and the Pod mitigation and audit steps to run now.
Read More
Kubernetes 1.36 security changes: user namespaces GA, fine-grained kubelet authz, gitRepo removed, externalIPs deprecated, and a Monday audit script.
Read More
Kubernetes 1.37 graduates manifest-based admission control to Beta: staticManifestsDir loads webhooks and CEL policies from disk, even when etcd is down.
Read More
How a fake Sentry error event became a command channel into Claude Code, Cursor and Codex, what the research does and does not prove, and the controls that hold.
Read More