eBPF
A technology in the Linux kernel that allows developers to safely run sandboxed programs in kernel space without modifying kernel source code. It is widely used in cloud-native observability, networking, and security tooling.
Core Facts
Timeline (last 90 days)
OpenShell's sandbox continuously intercepts and adjudicates the system calls and tool calls initiated by agents at runtime, similar to combining kernel-level interception mechanisms such as eBPF or seccomp with a user-space policy engine.
Keploy captures real traffic at the network layer via eBPF, records API calls and generates test cases and dependency mocks, without modifying application code.
eBPF is a virtual machine in the Linux kernel that can be dynamically attached to hooks such as networking, system calls, and tracepoints to execute sandboxed bytecode, and compared with iptables it offers higher programmability, lower performance overhead, and access to process-level metadata.
Using eBPF or network namespaces allows monitoring and restricting an agent's outbound connections, implementing a destination-address allowlist.
A more pragmatic self-hosted security route is to perform network isolation at the operating system or container level (such as eBPF or iptables/nftables), combined with a credential broker and short-term tokens, at the cost of increased deployment complexity.
A programmable UPF allows custom processing logic to be inserted into the data plane (such as based on P4 or eBPF), which is the landing point for IoD's implementation of in-band tensor context.
Truly effective network isolation requires traffic interception at the kernel level (such as using eBPF for mandatory auditing of outbound traffic), rather than relying solely on application-layer configuration constraints.
NetWatch's process attribution is achieved by combining platform-level socket polling, PKTAP on macOS, and optional eBPF kprobes on Linux.
eBPF provides multiple map types, including hash map, LRU hash map, and per-CPU map
A team successfully reduced the CPU cost of eBPF by approximately 90% by introducing memoization technology
10 more timeline events
All Facts (20)
eBPF was originally designed for network packet filtering and has since evolved into a general-purpose programmable infrastructure within the Linux kernel, allowing custom programs to be injected without modifying kernel source code; cloud-native security tools such as Cilium and Falco use eBPF as their core engine.
75%UnverifiedCaching frequently reused data in on-chip shared memory can reduce repeated reads from global memory.
70%UnverifiedeBPF allows developers to inject custom programs into the kernel for execution without modifying kernel source code; upon loading, programs are statically analyzed by the kernel verifier and then JIT-compiled into native machine code for execution, with performance approaching that of native kernel code.
60%UnverifiedOpenShell's sandbox continuously intercepts and adjudicates the system calls and tool calls initiated by agents at runtime, similar to combining kernel-level interception mechanisms such as eBPF or seccomp with a user-space policy engine.
50%UnverifiedKeploy captures real traffic at the network layer via eBPF, records API calls and generates test cases and dependency mocks, without modifying application code.
50%UnverifiedeBPF is a virtual machine in the Linux kernel that can be dynamically attached to hooks such as networking, system calls, and tracepoints to execute sandboxed bytecode, and compared with iptables it offers higher programmability, lower performance overhead, and access to process-level metadata.
50%UnverifiedUsing eBPF or network namespaces allows monitoring and restricting an agent's outbound connections, implementing a destination-address allowlist.
50%UnverifiedA more pragmatic self-hosted security route is to perform network isolation at the operating system or container level (such as eBPF or iptables/nftables), combined with a credential broker and short-term tokens, at the cost of increased deployment complexity.
50%UnverifiedA programmable UPF allows custom processing logic to be inserted into the data plane (such as based on P4 or eBPF), which is the landing point for IoD's implementation of in-band tensor context.
50%UnverifiedTruly effective network isolation requires traffic interception at the kernel level (such as using eBPF for mandatory auditing of outbound traffic), rather than relying solely on application-layer configuration constraints.
50%UnverifiedNetWatch's process attribution is achieved by combining platform-level socket polling, PKTAP on macOS, and optional eBPF kprobes on Linux.
50%UnverifiedA team successfully reduced the CPU cost of eBPF by approximately 90% by introducing memoization technology
50%UnverifiedHigh-overhead probes may distort the true performance characteristics of the observed system, i.e., the observer effect
50%UnverifiedeBPF provides multiple map types, including hash map, LRU hash map, and per-CPU map
50%UnverifiedFor large model API calls that go over HTTPS, network-layer observation can see the connection being established but the content is encrypted, requiring other means to reconstruct the call details.
50%UnverifiedAgentSight is a project that leverages eBPF technology to provide observability capabilities for AI agents without modifying any application code.
50%UnverifiedWell-known projects such as Cilium, Falco, and Pixie are all built on eBPF.
50%UnverifiedeBPF primarily relies on relatively recent versions of the Linux kernel, and its applicability in environments such as Windows and macOS is limited.
50%UnverifiedWhat is visible at the kernel layer consists of low-level signals such as system calls; mapping these back to the high-level decision semantics of agents requires additional correlation and interpretation work, presenting a semantic gap.
50%UnverifiedeBPF offers excellent performance but is limited to Linux, while ESF is feature-complete but requires special authorization.
50%