Protect your sshole before it gets penetrated
Security holes will not wait for patches. Design your web server so it holds up while they are still open: CDN, firewall, keys-only SSH and no root.
October and November were crazy in terms of security breaches and exposed vulnerabilites. AI is “helping” a lot to expose vulnerabilities. Most of these vulnerabilites were (still are) sitting for years laying down there in the code.
One of the latest security holes was named SSHOLE because of its relation to SSH. It allows attackers to can gain access by inputting more than 256 characters in the password field. It does not apply for all configurations and certains conditions have to be met but it is still a major security hole.
Other vulnerabilites that were discovered are related to NGINX and Caddy, NGINX being one of the most used web servers and reverse proxies.
So you have three ports that are almost always open on a linux web application server: 80, 443 and 22 (80 being used for redirects to 443 and SSH for remote management).
I come from infrastructure and for me a well designed, secure infrastructure is a building block for a secure environment, and because AI applications lives in this environment it is essential to think about the infrastructure before deploying AI, not after.
I prepared for you a list with recommended actions to improve the security on a web application linux server no matter if it is AI enabled or not.
- Use CDN - A CDN (content delivery network) is a worldwide set of servers that sits between your visitors and your server, serving copies of your pages from many locations. It protects your server because attack traffic is absorbed and filtered by the CDN first, so your real server is never exposed directly to the internet. I use CloudFlare but there are different options
- Configure your firewall to accept connections on port 80 and 443 only from the CDN. All CDN has a bunch of IP addresses that you can whitelist so that only traffic from them will reach your server on these two ports. Depending on the setup you can do that on your WAF (Web Application Firewall) or directly on the server if your application is hosted only one server. And if you are using VPS you can probably also configure these rules on the host provider so that connection dont even reach your server but the host infrastructure.
- NEVER EVER USE ROOT ACCOUNT This should be the first thing to do when you build a server. Create a new account (or accounts if it will be used from multiple people) that can elevate permissions when necessary and disable the root account. First - root is the first target. It has unlimited power over the whole server, and automated attacks try it first. If root cannot log in, an attacker must first steal a normal account and then escalate privileges. Second - when everyone logs in with their own account and uses sudo, every administrative action is traced to a named person in the logs. A shared root login hides who did what. Third - It limits the damage of a stolen password. With the root password locked, a leaked password for root cannot open the server, through SSH, su or the console.
- Use keys to login via SSH and disable password logins SSH allows you to login with passwords and keys. By disabling password login you decrease the chances of attacks significantly. Keep keys per machine. If you are logging from different machines use different keys for each of them so that if a machine is compromised you can just disable the selected keys.
- BACKUPS, BACKUPS, BACKUPS! Make backups regularly and keep a second copy on an offsite location. Make sure the backup files are also encrypted and the server cannot read the encrypted files. This will ensure that if an attacker gains access to your server and try to recover something from backups he will not have access to them.
- Patch your systems As you saw the last two months were crazy and I don’t see security researchers slowing down with discovering vulnerabilites. Software vendors cannot keep up with patching all of them ….. OpenAI reported that an automated system found 948 distinct Linux privilege-escalation flaws and wrote working exploits for them (Decrypt, 5 November 2026)
These are only basic precautions that apply for almost any web application server. Depending on the application and its features that you are publishing there other recommendation that you might want to consider but they depend on the technologies that you will be using. Consider what scripts are running on your website, what input fields are accessible to the user (contact forms, login forms, AI chat, etc), what API calls are being made to the server. Each form of request being sent to your server presents different attack surface and limiting its exposure by design is key to stay ahead of the zero day game.
Security is not telling Claude “Making my application secure.” This will only make you feel pretty but the foundation of your app will stay roted. Security is a continuous effort not a one time thing that you ask your AI coding agent.