Security infrastructure itself is under pressure

F5 / KEV / access

F5 BIG-IP APM flaw is being exploited and lands on KEV with a three-day fuse

CISA added CVE-2026-94127 to the KEV catalog on September 22. CISA describes a heap-based buffer overflow in BIG-IP APM that occurs when an access policy and an OAuth profile are configured on a virtual server, allowing an unauthenticated attacker to achieve remote code execution. F5 has reported exploitation in the wild, and the Canadian Centre for Cyber Security lists hotfixes for BIG-IP APM 21.1.0, 17.5.0 and 17.1.0 branches.

Why it matters. APM is the access gateway — it terminates authentication for the applications behind it. A pre-authentication bug there does not just compromise a box; it sits in front of the identity decisions for everything it fronts, which is why CISA set a September 25 due date and requires forensic triage under BOD 26-04.

CVE-2026-94127

CISA — CISA Adds Four Known Exploited Vulnerabilities to Catalog|CISA — Known Exploited Vulnerabilities Catalog|F5 — K000162605|Canadian Centre for Cyber Security — F5 security advisory AV26-949

Check Point / Arista / KEV

Check Point and Arista management planes join the same KEV batch

The same September 22 KEV addition covers Check Point CVE-2026-85102, an improper certificate validation flaw in Security Gateway and Spark firewalls using site-to-site or remote-access VPN; Check Point CVE-2026-93616, a pre-authentication path traversal in Security Management; and Arista CVE-2026-93952, improper input validation in on-prem VeloCloud Orchestrator. Check Point says a fix for CVE-2026-85102 has been available since September 9 and that it is now seeing exploitation attempts against Spark customers, while CVE-2026-93616 was a zero-day with a fix released with the advisory. Arista says its flaw "was discovered externally and is known to be actively exploited," scores it 10.0 on CVSS 3.1, and notes only certificate-authenticating deployments are exposed.

Why it matters. These are not ordinary application bugs. A firewall management server, an SD-WAN orchestrator and a VPN gateway each hold authority over many downstream devices, so one compromise inherits a fleet. Arista's own advisory notes a compromised orchestrator can reach the edges it manages.

CVE-2026-85102, CVE-2026-93616, CVE-2026-93952

CISA — CISA Adds Four Known Exploited Vulnerabilities to Catalog|Check Point — Security advisory on CVE-2026-85102 and CVE-2026-93616|Check Point — sk1000117|Arista — Security Advisory 0144

AI / e-commerce / threat intel

Researchers report an AI-agent campaign stealing payment cards from online retailers

Gambit Security reports that a financially motivated operator has been running three open-source AI harnesses — Strix for vulnerability discovery, Cairn for exploitation and Hermes for orchestration — against online retailers since July 2026, largely unattended. Gambit says it reconstructed the campaign from the operator's exposed staging server and accounts for more than 600,000 unexpired payment card records taken from two companies, skimmer scripts installed on five retailer websites, and 105 attack projects launched between September 10 and 15 with at least 27 organizations compromised to some degree. Gambit puts the operator's own reported mean cost at $25.46 per completed scan.

Why it matters. These figures are researcher-reported from one recovered server, not an independently audited tally, and are worth reading as such. The economics are the point regardless: if a competent intrusion costs tens of dollars and runs without an operator watching, the filter that kept mid-sized retailers below an attacker's attention threshold stops working.

Gambit Security — Autonomous AI agents are breaking into hundreds of online retailers|CyberInsider — AI agents steal 600,000 credit cards in attacks on online retailers

supply chain / Terraform / Go

Graphalgo malware turns up in Terraform providers and Go modules

Aikido Security reports Go malware distributed through two Terraform providers, gocommunity-io/dockerd and kreuzwenker/docker — the latter a typosquat of kreuzwerker/docker — and two Go modules, gocommunity.io/orderedbtree and gogets.dev/btreex. Aikido says the code only activates when specific Terraform variable values hash to a hardcoded value, and links the malware to the Graphalgo npm campaign through shared blockchain and Slack infrastructure and a shared public key. Aikido advises treating any version of those four packages as an indicator of compromise. Reporting on Aikido's findings notes 18 distinct hostnames observed in check-ins, which does not establish how many organizations were affected.

Why it matters. Terraform providers run during plan and apply with whatever cloud credentials the pipeline holds. Malware there is not on a developer laptop, it is inside the process that creates and authorizes infrastructure. The trigger-gated design also means ordinary testing looks clean.

Aikido Security — Graphalgo campaign spreads to Terraform providers and Go modules|Cyber Security News — Malware hiding in developer tools