WAF Weekly: exploited edge and CI/CD flaws, AI agents, Sep 12-18
Cisco patches an exploited email gateway zero-day, GitLab fixes a maximum severity CI/CD flaw, scanners chase Vite dev servers, and Anthropic details AI-orchestrated cyber operations.
Exploited edge flaws and a rising AI agent threat defined the week. Cisco patched an in-the-wild Secure Email Gateway zero-day that grants root, GitLab fixed a maximum severity CI/CD flaw, and automated scanners hammered exposed Vite dev servers for cloud credentials. Anthropic also detailed how AI now orchestrates state-linked cyber operations.
Cisco Secure Email Gateway zero-day exploited in the wild
Cisco shipped emergency patches for CVE-2026-76461, an SQL injection in the Secure Email Gateway's email parsing code that lets an unauthenticated attacker run commands as root by sending a crafted message. The flaw was already under active exploitation when fixes landed, and CISA added it to its Known Exploited Vulnerabilities catalog. Because a successful attack means root access, attackers can alter device logs to hide their tracks, so Cisco advises reviewing external firewall and network logs.
The WAFplanet angle: email gateways are edge appliances that parse untrusted traffic by design, which makes them prime targets. A root-level, unauthenticated RCE here is a durable foothold. Treat edge appliances as assets worth defending and do not rely on on-device logs alone after a suspected exploit.
Maximum severity GitLab flaw exposes CI/CD servers
CVE-2026-85706 is a perfect-10 path traversal in GitLab's repository commits API that lets an unauthenticated attacker read arbitrary files, including secrets and credentials, in a single HTTP request. It is the second critical GitLab flaw in a month, and watchTowr reports in-the-wild probes are already underway. Self-hosted CE and EE instances are affected, and GitLab sits at the center of build and deploy workflows, so a file read there can cascade into supply chain compromise.
The WAFplanet angle: CI/CD platforms are critical trust infrastructure, and GitLab has become a favorite target. Unauthenticated file access on a platform connected to pipelines and clouds is worse than a typical web app bug. Patch immediately, audit repository commits API traffic for file.path parameters, and treat build platforms as attack surface worth an edge WAF in front of.
Scanners target exposed Vite dev servers for cloud credentials
F5 Labs documented a sharp jump in scanning of exposed Vite development servers. Attacks on CVE-2026-39364, a file-read bypass of the server.fs.deny deny-list, climbed from 1,732 events over the prior three months to more than 32,000 in August. The scanning fleet cycled through wordlists for .env files, AWS keys, Azure tokens, and Infrastructure-as-Code state files, using double-encoded path traversal plus forged crawler agents to slip past reverse proxies and WAFs.
The WAFplanet angle: development tooling is the new supply chain soft spot. These asks hit a dev server accidentally mapped to the internet, not a published endpoint. Keep dev and preview ports off external interfaces, and WAF rules tuned to deny-list bypass patterns earn their keep.
WordPress plugins under active attack
Threat actors are actively exploiting CVE-2026-27540 in WooCommerce Wholesale Lead Capture, a premium plugin with more than 6,000 installs. The arbitrary file upload lets unauthenticated attackers plant PHP web shells for remote code execution, and Wordfence says it has blocked over 100,000 exploit attempts since June. The same research detailed two critical unauthenticated RCE chains in The Events Calendar, installed on over 600,000 sites, where pending-comment preview can trigger full site takeover.
The WAFplanet angle: plugin ecosystems remain the dominant WordPress risk, and active exploitation means patching alone is not enough. Check for unexpected .php files in uploads directories, monitor admin-ajax.php traffic, and a WordPress-specific WAF that inspects plugin and AJAX requests is a practical layer between attackers and core files.
Anthropic details AI-orchestrated cyber operations
Anthropic's September threat intelligence report covers operations it disrupted from December 2025 through August 2026, including a Russian espionage actor linked to Midnight Blizzard. The actor automated reconnaissance, phishing, persistence, and exfiltration with custom AI-driven workflows, and rebuilt its malware autonomously when security products flagged it. Anthropic's core finding is that AI has collapsed the gap between well-resourced states and individual operators, with sophisticated attacks no longer requiring sophisticated attackers.
The WAFplanet angle: the detect-and-signature cycle breaks when an adversary can modify tooling the moment it is flagged. This shifts weight toward behavioral and intent-based controls. Expect more AI-orchestrated phishing and abuse traffic, and lean on per-request verification and edge filtering that does not depend on known signatures.
Also notable
- Hundreds of OpenAI agents uploaded malicious packages and probed for API keys on RubyGems, prompting questions about guardrails and accountability (CSO Online)
WAFplanet take
This week shows the perimeter widening faster than defenses are re-architected. The common thread across the most dangerous incidents is unauthenticated access to infrastructure that sits closer to secrets than browsers do, email gateways, CI/CD platforms, and dev servers. Meanwhile the AI agent wave automates credential and package attacks at scale while making it easier to misattribute hostile traffic as benign automation. Edge protection that filters on behavior, not reputation or known signatures, keeps value against both trends.
We also read
- Wordfence launches free passkeys to secure WordPress logins (Wordfence)
- Wordfence Argus identifies two critical unauthenticated RCE chains in The Events Calendar (Wordfence)
- Wordfence publishes its Bug Bounty Program monthly report for May 2026 (Wordfence)
- Fastly shares a new DDoS playbook from Khan Academy, Yottaa, and Magnolia (Fastly)
- Prophaze on WAF features for compliance-ready government agencies (Prophaze)
- Fortinet named a Leader in the 2026 Gartner Magic Quadrant for Hybrid Mesh Firewall (iTWire)
