Explore our new Cybersecurity Terms Glossary, explained by our experts.

View the Glossary View the Glossary
PRIVIA

Cybersecurity Terms Glossary

What Is Falco?

Falco watches kernel system calls to detect unexpected behaviour inside containers and Kubernetes clusters at runtime, generating alerts from rules rather than from file signatures.

Illustration for the cybersecurity term Falco
Home / Blog / Cybersecurity Terms / What Is Falco?
17 November 2026

What Is Falco?

What Is Falco?

Falco is an open-source runtime security tool for containers, Kubernetes, and Linux hosts. It attaches to the kernel, observes the system calls that processes make, enriches that stream with container and Kubernetes metadata, and evaluates it against a rule set. When behaviour matches a rule, it emits an alert.

It was created by Sysdig in 2016, contributed to the Cloud Native Computing Foundation in 2018, and became a CNCF graduated project in February 2024, which put it in the same tier as Kubernetes and Prometheus. For many organisations it is the first runtime detection they deploy in a cluster, largely because the alternative is nothing.

The Gap It Fills

Most container security effort happens before anything runs. Image scanning finds vulnerable packages, policy engines such as OPA Gatekeeper and Kyverno reject non-compliant manifests at admission, and Pod Security Standards constrain what a workload may request.

All of that is preventive, and all of it stops at the moment the container starts. If an application is compromised through a flaw that no scanner knew about, the container is running an approved image, from an approved registry, with approved settings, doing something nobody intended. Runtime detection is what notices.

The behaviours that matter are also unusually easy to specify in a container context, because containers are supposed to be predictable. A web server container should never launch a shell. A container should not write to /etc. A process inside a container has no business reading /etc/shadow, loading a kernel module, or opening a raw socket. In a general-purpose server, defining normal is hard. In a single-purpose container, it is nearly trivial, and that is why runtime rules work better here than almost anywhere else.

How It Gets the Data

Falco originally used a kernel module. The modern default is eBPF, which runs a verified program in the kernel to capture system call events without loading custom kernel code, and a CO-RE build removes the need to compile against each kernel’s headers. A plugin framework extends the same rule engine to event sources that are not system calls, including the Kubernetes audit log, AWS CloudTrail, GitHub audit events, and Okta logs.

Rules are YAML. Each has a condition expressed over event fields, a priority, and an output template. A simple example in plain terms: alert when a process whose name is a shell starts, and its parent is a container, and the image is not on the exception list. Falco ships with a maintained default rule set covering terminal access to containers, writes below sensitive directories, unexpected outbound connections, privilege escalation attempts, contact with the cloud instance metadata service, and mounting of sensitive host paths. The rules map to MITRE ATT&CK techniques in the ATT&CK Containers matrix.

Output goes to a sidecar called Falcosidekick, which forwards to Slack, Elasticsearch, a SIEM, or a webhook, and the Falco Talon component can take automated response actions such as terminating or isolating a pod.

Making It Work in Practice

Falco deployed with default rules and no tuning produces a great deal of noise, and the usual outcome is that the alerts are muted and the tool is quietly abandoned. Three practices avoid that.

Tune before you connect it to anyone’s attention. Run it in a non-production cluster for a fortnight, collect what fires, and add exceptions for the legitimate cases. Debugging shells in development namespaces and package managers running in build containers are the usual top two.

Treat rules as code. Keep custom rules in version control, review changes, and test them against recorded event captures rather than editing them live.

Write rules for your own applications. The default set is generic. The valuable rules are specific: this payments container runs exactly three executables, and anything else is an incident.

Where It Sits Among Alternatives

Tetragon, built on eBPF by Isovalent, offers in-kernel enforcement as well as observation. Tracee from Aqua covers similar ground. Commercial cloud workload protection platforms bundle runtime detection with posture management and vulnerability data in one console. Falco’s advantages are that it is free, vendor-neutral, widely understood, and well documented, which makes it a reasonable starting point even for organisations that will later buy a commercial product.

One limitation to plan for: Falco needs privileged access to the node kernel, which means it is generally unavailable on managed serverless container platforms such as AWS Fargate, where the provider controls the host.

At Privia Security, our Cloud Security Services include validating that container runtime detection actually fires against the techniques an attacker would use after compromising a workload.

FAQ – Frequently Asked Questions

Does Falco Block Attacks or Only Alert?

Falco itself detects and alerts. Response is handled by companion components: Falcosidekick forwards events to messaging, logging, and SIEM destinations, and Falco Talon can take automated actions such as terminating or isolating a pod. For in-kernel enforcement at the point of the system call, tools such as Tetragon are designed for that role.

How Is Falco Different From Image Scanning?

Image scanning inspects what is inside a container before it runs and reports known vulnerable packages. Falco observes what the container actually does once it is running. A compromised application in a fully patched image passes every scan and still spawns a shell, and only runtime detection sees that. The two controls cover different halves of the problem.

Will Falco Slow Down Our Cluster?

The overhead is modest with the eBPF driver, typically a few percent of CPU on the node, but it scales with system call volume. Workloads performing very heavy file or network input and output generate the most events. Measure it in your own environment, and tune noisy rules, since rule evaluation on a high-volume event stream is where most of the cost appears.

What Should We Do About Alert Noise?

Tune before anyone is expected to act on the output. Run Falco in a non-production cluster for a couple of weeks, review what fires, and write exceptions for legitimate cases such as debugging shells in development namespaces and package managers in build containers. Then add narrow rules for your own workloads, which produce far higher-value alerts than the generic set.

Related Terms

You May Be Interested In These

What Is Netcat?

Cybersecurity Terms

17 November 2026

What Is Netcat?

Read More
What Is IPsec?

Cybersecurity Terms

17 November 2026

What Is IPsec?

Read More
What Is Threat Hunting?

Cybersecurity Terms

17 November 2026

What Is Threat Hunting?

Read More