A defender-focused analysis of how APT29-aligned and Storm-2372-style intrusion sets are abusing device authorization grants and long-lived OAuth consent in Entra ID through 2026 — what to detect, what to revoke, and where it maps to 800-53.
A defender-oriented analysis of firmware- and bootloader-resident implants on perimeter network appliances, what they look like in practice, how to detect them, and how to map the response to NIST 800-53.
Hyunwoo Kim’s Dirty Frag chain extends the Dirty Pipe / Copy Fail class to skb paged fragments. The xfrm ESP receive path provides a deterministic 4-byte page-cache store (CVE-2026-43284); the rxrpc receive path provides a namespace-free trigger (CVE-2026-43500). One PoC, no race, root on Ubuntu, RHEL, CentOS Stream, AlmaLinux, Fedora, and openSUSE — including hosts that already blocklisted algif_aead for Copy Fail.
Immutable base images plus systemd sysext/confext overlays have quietly become the standard for fleet-managed Linux in 2026. They also created a persistence surface most SOCs are not yet watching.
A double-free in Apache httpd 2.4.66’s mod_http2 lets unauthenticated clients crash workers reliably and, on certain builds, opens a path to remote code execution. Here’s the defender-oriented breakdown: mechanism, detection, remediation, and 800-53 mapping.
Model Context Protocol gives agents a clean way to call tools, but it also gives every piece of retrieved content a direct line to those tools. The classic confused deputy problem is back, and most deployments are not handling it.
Model Context Protocol turned every internal API into an LLM-reachable tool. The threat model didn’t catch up. Here’s what tool poisoning looks like in production and how to harden against it.
Theori’s Copy Fail flaw turns a nine-year-old in-place optimization in the kernel’s AEAD socket layer into a deterministic 4-byte page-cache write. No race, no offset hunting, no per-distro tuning. A 732-byte Python script edits a setuid binary’s cached pages and hands you root on Ubuntu, RHEL, SUSE, and Amazon Linux without modification.
Indirect prompt injection has graduated from a research curiosity to a routine cause of agent compromise. Here is what defenders are actually shipping in 2026, and where the control gaps still are.
Agentic AI breaks the assumptions baked into service accounts and long-lived API keys. Here is what an honest non-human identity model looks like when an LLM is the principal making the call.
