Blog

This week in web attacks: WordPress upgrade flaw, Apache HTTP/2 and OpenSSH

This week one day dominated our numbers, and the wider news was full of flaws that let attackers run code on web servers. Several were already being exploited. Here is what we saw, what others reported, and what is worth doing before the weekend.

What we saw on our sites

All figures here come from the 16 sites DBshieldX protects, counted together across our servers.

Over the week we inspected 191,157 requests and recorded 32,983 detections. Of those, 29,088 were blocked. A further 303 were flagged on sites running in monitor mode, where a request is logged but not stopped. Those 303 are requests that would have been blocked in block mode.

The week was very uneven. On 16 November we recorded 24,418 detections, with 23,950 blocked, out of 44,123 requests inspected. That is more than half of that day’s traffic. The other seven days we have counts for ranged from 787 to 3,190 detections, with 13 November the next highest at 3,190.

By class, the rule groups that matched most often were:

  • Behaviour (suspicious request patterns): 26,580
  • Path (probing for files and paths): 15,799
  • Bot (automated scanners and bad bots): 10,226

The injection-style classes were much smaller. We counted 735 exploit, 732 command injection, 383 SQL injection, 96 server-side request forgery, 88 cross-site scripting, 33 LDAP, 16 NoSQL and 4 template injection detections. The class counts add up to more than the detection total, so one request can be tagged under more than one class.

The traffic was concentrated. About 85% of detections (28,105) came from servers hosted on Google Cloud, and roughly seven in ten (22,857) came from servers located in Singapore. Hong Kong-based servers accounted for 1,605, and servers hosted by UCloud Information Technology (HK) for 1,377. Anyone can rent servers at these providers, and a country only tells you where a server sits, not who is behind it. The same burst may explain both the cloud and country figures, but our counts don’t prove that.

What others reported

  • WordPress core upgrade flaw. Wordfence says attackers are probing for CVE-2026-119977, a code execution flaw that can be triggered without logging in while WordPress is upgrading. It affects versions 4.7 to 7.1.3, and the fix is 7.1.4 or the backported release for your branch. The CISA-listed entry points to predictable staging paths that can be reached over the web. The Hacker News adds a catch: installing the fix can leave some sites exposed one last time, because the update still unpacks files using old code. The researcher advises blocking web access to the update path while you update.
  • Apache HTTP/2. A critical flaw in Apache’s HTTP/2 module can let an unauthenticated attacker run code remotely. The Hacker News says it affects versions 2.4.55 to 2.4.70, and only servers with HTTP/2 enabled. The fix is 2.4.71, or you can switch off the module if you don’t need it. Wordfence explains what it means for WordPress sites, and BleepingComputer covers the patch.
  • OpenSSH. BleepingComputer reports that attackers are exploiting CVE-2026-123158 to run commands, steal credentials and keep access to Linux systems. The NVD entry notes it is in CISA’s Known Exploited Vulnerabilities catalogue. Most web hosts rely on this service for admin access.
  • React Server Components. A GitHub advisory describes a critical flaw that can allow unauthenticated code execution in rare cases. Frameworks that depend on the affected packages, such as Next.js, need updating to the patched versions the advisory names.
  • Rails Active Storage. Another GitHub advisory says Rails apps that process uploads with ImageMagick can let an unauthenticated attacker write arbitrary files. With disk storage they could also read sensitive files, which can lead to code execution.
  • AI gateway apps. Amazon observed automated agents probing internet-exposed AI gateway tools such as LiteLLM and Dify to steal stored provider API keys. Separately, Railway says stolen customer AI keys led to an estimated $740,000 in unauthorised charges.

How this relates

Our counts don’t say which specific flaw any of our detected traffic was after, so we won’t link the two directly. What we can say is that the most common kinds of detection, probing for paths and suspicious request patterns, are the groundwork that comes before exploitation of web-reachable weaknesses. The WordPress upgrade flaw is a good example: it depends on paths that can be reached from the internet.

One day of heavy traffic from a handful of rented servers is also a reminder that a firewall layer such as DBshieldX can cut down noise and block known bad patterns using rules. It does not replace patching, and no rule set catches everything. Several of this week’s flaws are in the software underneath your site, and the real fix for those is the update.

What to do this week

  1. Update WordPress to 7.1.4 (or the backported release for your branch). Block web access to the update and staging paths while you do it, then check for unexpected files afterwards.
  2. Check Apache. Ask your host whether 2.4.71 is installed. If you manage the server and don’t need HTTP/2, you can disable the module until you can upgrade.
  3. Patch OpenSSH on every server. Restrict SSH to trusted addresses or a VPN, rotate keys and passwords on exposed hosts, and review recent logins and authorised keys for anything you don’t recognise.
  4. Update framework dependencies. If you run Next.js or other React Server Components frameworks, or Rails with Active Storage and ImageMagick, apply the patched versions named in the advisories.
  5. Take AI admin panels off the public internet. Put them behind an IP restriction or VPN, change default credentials, and rotate any provider API keys stored in those tools.
  6. Review your logs. Look for odd requests to upgrade or staging paths and for files you didn’t create. If you find signs of compromise, rebuild the host rather than only patching it.

See it on your own traffic.

We'll run DBshieldX in monitor mode on one of your sites for a week, free, then walk you through what it would have blocked. Nothing is blocked until you say so.