WAF Weekly: FortiWeb patches, HTTP/2 Bomb, Signal key transparency, Aug 10-14

This week in web security swung between patches and persistence. Fortinet closed an authentication hole that could hand attackers the keys to its FortiWeb WAF, a new HTTP/2 compression bomb took down web servers in seconds, and Microsoft shipped 421 fixes including a Windows zero-day already under attack. Meanwhile Cloudflare moved deeper into trust infrastructure and researchers documented the first near-autonomous AI agent attack on government networks. For teams running WAFs, the through-line is simple: keep the software current, because attackers are automating the window between advisory and exploit.

Fortinet patches FortiWeb authentication bypass

Fortinet fixed a high severity flaw in FortiWeb that let a remote, unauthenticated attacker log into the WAF's GUI or CLI with a random username and password. Tracked as CVE-2026-26035, the issue only bites deployments where the wildcard setting for administrator accounts is enabled, which is off by default. When enabled, the system matches any username on a remote server against the defined admin group. Patched versions include 8.0.3, 7.6.7, 7.4.12 and 7.2.13, with a disable-wildcard workaround if you cannot patch. This one matters because a WAF that an attacker can log into is not a control, it is a beachhead. If you run FortiWeb, verify whether wildcard auth is on and patch now.

Read the Fortinet advisory coverage at SecurityWeek

HTTP/2 Bomb knocks web servers offline in seconds

Security researchers at Calif chained old HTTP/2 techniques into a new DoS exploit that can take down servers in seconds from a home connection. The HTTP/2 Bomb pairs a header compression attack with a Slowloris-style memory hold, hitting NGINX, Apache HTTPD, Microsoft IIS, Envoy and Cloudflare Pingora. Roughly 880,000 sites are potentially exposed. NGINX and Apache have patched, while IIS, Envoy and Pingora had not at the time of writing. Cloudflare said its architecture detects the attack automatically. Notably, the exploit was assembled by an AI coding agent, Codex. WAFs and reverse proxies are your first line here, so review your vendor's byte and connection limits.

Read the HTTP/2 Bomb coverage at SecurityWeek

Cloudflare and Trail of Bits audit Signal key transparency

Signal shipped automatic key verification this week and brought in two independent auditors, Cloudflare and Trail of Bits, to sign the tamper-evident log that keeps public keys globally consistent. The idea is to catch a key swap before it breaks end-to-end encryption. Cloudflare built the system in Rust on its Workers platform and already runs the same infrastructure for WhatsApp. The independent auditor design means no single party can show one log to one user and a different one to another. This is the same verification problem that every encrypted channel faces, and it is worth watching how provable trust becomes a default standard rather than a feature.

Read the key transparency coverage at The Next Web

Open-source AI agents breach Taiwan government networks

Researchers at Dream documented a four-day near-autonomous attack in which AI agents built on open-source frameworks breached Taiwanese government systems, cracked 85 credentials and probed a nuclear safety agency. The system ran multiple sub-agents in parallel across 12 attack waves, mapping networks, targeting unauthenticated APIs and bypassing token validation, then moving laterally through single sign-on. Taiwan's Ministry of Digital Affairs confirmed an AI-assisted attack in the same window. The report does not confirm a full breach of the nuclear body. The headline is the collapse in attack cost, and the warning for defenders is that automated reconnaissance now finds the gaps you have not patched.

Read the AI agent attack coverage at CSO Online

Microsoft August Patch Tuesday fixes 421 bugs

Microsoft's August update covers 421 vulnerabilities including the Windows zero-day CVE-2026-68820, a use-after-free in the afd.sys kernel driver that lets attackers gain SYSTEM privileges and is already exploited in the wild. Another Windows User Profile Service elevation bug and a Container Isolation tampering flaw were publicly disclosed before patching. Windows accounts for 236 of the fixes with Office next at 98. There is also a potentially wormable Windows DNS Server RCE and a QUIC flaw affecting millions of sites. Patch Tuesday is a reminder that the web-facing surface and the kernel surface are the same inbound attack plane, and both need WAF plus patching discipline.

Read the Patch Tuesday coverage at TechnoBezz

Also notable

• VMware patched vCenter directory traversal CVE-2026-59310, now exploited in the wild with reverse shell payloads across hundreds of victim IPs.
• WordPress 7.0.4 fixed a remote code execution vulnerability, the kind that web application firewalls are built to blunt.
• A patched SharePoint vulnerability is now being exploited in the wild shortly after a proof of concept went public.

WAFplanet take

Two patterns dominated this week. First, the patch window is shrinking. VMware, SharePoint and a Windows zero-day all went from disclosure to active exploitation within days, and AI agents are doing the reconnaissance that used to take skilled humans. Second, the boundary between web security and infrastructure security is dissolving, because an unpatched vCenter or a FortiWeb login is the same entry point as a classic WAF bypass. A WAF is not a substitute for patching, it is the net that catches what the patch race misses. Keep your WAF signatures and versions current, treat every admin interface as a live target, and assume an automated attacker is already scanning what you have left exposed.