Threat Brief — 2026-09-19 — Three More Linux Kernel CVEs Hit KEV

CISA has added three additional Linux kernel vulnerabilities to its Known Exploited Vulnerabilities catalog, spanning the TLS receive path, ebtables SNAT, and AF_ALG sockets — all confirmed as exploited in the wild. Separately, a new analysis highlights that the AI tooling layer is accumulating CVEs at a rapid pace, with the vast majority of integrations lacking OAuth authentication, leaving agent and plugin ecosystems broadly exposed.

Top items

• Three Linux kernel CVEs newly listed in CISA KEV (exploited in the wild). CISA added CVE-2025-39682KEV (improper check in the TLS receive path allowing zero-length record handling), CVE-2026-53266KEV (out-of-bounds write in the ebtables SNAT target via ARP sender hardware address rewrite), and CVE-2025-39964KEV (race condition in AF_ALG sockets allowing concurrent writes to produce unpredictable data interleaving and socket inconsistency). All three are Linux kernel flaws now confirmed as actively exploited. This continues a story first reported 2026-08-17 by CISA current-activity alerts; the catalog continues to accumulate Linux kernel entries. (src: CISA KEV)

• AI tool layer accumulating CVEs with minimal authentication. An analysis reports 68 CVEs discovered in AI tooling over a single month, with 91.8% of integrations operating without OAuth authentication. This compounds recent reporting on AI-agent security gaps — default configurations, plugin-swap attacks, and prompt-injection credential theft — suggesting the AI tooling ecosystem is growing faster than its security controls. (src: Anquanke)

Themes

Linux kernel KEV momentum. CISA's KEV catalog has now absorbed multiple Linux kernel vulnerabilities across distinct subsystems (TLS, ebtables, AF_ALG) within a short window, signalling broad exploitation of kernel-level flaws rather than a single targeted campaign. Organisations running Linux-based infrastructure should treat kernel patching as time-critical when KEV listing occurs.

AI tooling as an emerging attack surface. The convergence of this CVE analysis with recent reports of plugin-swap attacks, prompt-injection credential theft, and AI-assisted malware development points to a maturing threat landscape around AI agent ecosystems — one where authentication, plugin integrity, and sandboxing are lagging behind adoption.