Egress Filtering

Every corporate network and every data center has some kind of boundary to the open internet. At this boundary usually sits a gatekeeper program that inspects traffic. Most people only think of this in one direction: what is allowed to come in from outside? Egress filtering turns the question around and checks what is allowed to go from inside to outside. "Egress" is the English word for exit. A machine on the network is then only allowed to talk to destinations that are explicitly permitted — everything else is blocked.

Why attackers need a way back out

Breaking into a network is only half the job for an attacker. Stolen customer data is useless to them as long as it sits on the machine inside the corporate network. They have to get it out somehow, usually via an internet connection to a server they control. This is exactly the connection that egress filtering blocks. The technical term for smuggling data out is exfiltration.

The same applies to malware that has already established a foothold. Such programs regularly check in with a control server on the internet and fetch new commands from there. This is called command-and-control traffic. Without this return channel, the malware is largely paralyzed. It may sit on the network, but it can neither be controlled nor deliver anything.

That is why security agencies and auditors now treat outbound control as mandatory practice. Many incident reports show that the actual breach happened weeks earlier. It was only noticed when an unusually large amount of data left the network. So whoever monitors the exit gains not only protection, but also a chance at early detection.

From blocklist to allowlist

Technically, filtering works with rules stored on a firewall or a proxy. A firewall is the program that inspects data packets at the network boundary and either lets them through or discards them. Each rule describes who may talk to whom and over which channel. Such channels are called ports and are numbered: port 443, for example, stands for encrypted web pages.

What matters is the underlying stance of the rules. A blocklist allows everything and blocks individual known offenders — convenient, but full of holes. The secure variant is the allowlist: everything is forbidden unless explicitly listed. A database server usually needs no internet access at all. It simply gets none, and an entire attack path disappears.

In practice this is more demanding than it sounds. Attackers like to hide their data in traffic that looks harmless, such as DNS requests — these are the requests a machine uses to look up the address of a website. Almost every network has to allow these requests. Modern systems therefore check not only the destination, but also the volume and pattern of traffic. A server that suddenly sends out twenty gigabytes at night triggers an alert.

Where the rule shows up in everyday life

On your own school network or company Wi-Fi, you often notice outbound control through small annoyances. An online game won't connect, a messenger stays silent, a certain app loads forever. Usually this is because these programs use unusual ports that simply aren't permitted. That is exactly the downside: overly strict rules also block legitimate work.

This topic has become especially important with cloud services and AI systems. When a company runs a language model on its own servers, it wants to be sure that no internal documents flow out unintentionally. AI agents that carry out tasks independently on the internet also get tight outbound rules. They are then only allowed to contact a short list of vetted addresses.

In news reports about data leaks, the term usually appears in hindsight. It is then said that there was insufficient control of outbound traffic. A common misconception, incidentally, is confusing egress filtering with encryption. Encryption protects data in transit from eavesdroppers. Egress filtering, by contrast, decides whether the data is allowed to leave at all.