eBPF for Security: Listening to Your Kernel in Real Time
How six programs running deep inside Linux detect privilege escalation, block malware, and drop malicious packets instantly.
Kernel hooks, syscall-level detection, firmware checks, correlation rules, Sigma translation, anomaly detection, and production lessons.
How six programs running deep inside Linux detect privilege escalation, block malware, and drop malicious packets instantly.
From 7 monitors to 22. Container escapes, hidden malware, and rootkits: three real attack scenarios detected deep in the system, with smart noise filtering.
How to detect remote access attacks by tracking system behavior instead of pattern matching. Impossible to evade via obfuscation.
Why regex fails for obfuscated commands like hex-encoded payloads, base64 pipelines, and Python reverse shells. How tree-sitter AST analysis detects them structurally.
How we imported 208 community detection rules and adapted them for real-time kernel monitoring. Rewrites, false positive fixes, and practical results.
Real lessons from running 40 eBPF hooks (tracepoints, kprobes, LSM, XDP). #![no_std], CO-RE/BTF, ring buffer with epoll, common verifier failures, debugging.
Deep dive on JA3/JA4 TLS fingerprinting in pure Rust with AF_PACKET. 10 known malicious hashes, GREASE filtering, and how to add custom fingerprints.
47 rules correlating events across Ring -2 firmware, Ring 0 kernel, userspace, network, and honeypot layers. How to detect multi-stage attacks that no single product can see.
All 69 cross-layer correlation rules. Firmware chains, network chains, execution chains, post-compromise patterns. Each with attack scenario and time window.
Secure Boot, TPM, ESP hashing, UEFI variable tracking, ACPI table scanning, and boot timing anomalies. Six checks that catch BlackLotus, LoJax, and MosaicRegressor before the OS loads.
Using CPU timing analysis to find hidden rootkits without any kernel module. Based on published security research with 98.7% accuracy.
7 days of training, then anomaly detection without rules. Process lineage anomalies, silence detection, login time deviations, and unknown network destinations.
Why we abandoned our previous AI model and built a tiny, fast anomaly detector in Rust. It learns what normal looks like on your server and catches anything unusual.
130 Telegram alerts overnight. 62% were false positives. How we fixed each one by understanding what normal looks like, not by tuning thresholds. Six real examples with code.
Sigma + YARA + IDS rules vs heuristics vs baseline learning. Honest trade-offs, why a layered approach wins, and how the autoencoder + Sigma + correlation engine layer up.
From the deepest hardware layer to user applications, in one Rust binary. 40 monitors, 82 detectors, 69 correlation rules, behavior learning, mesh network. The full picture.
What Inner Warden sees that nobody else does: deep hardware monitoring, firmware threats, and hidden attacks. A factual gap analysis.
5 rounds of adversary emulation with Caldera v5.3.0. From 36% detection to 67% of testable techniques. The argv truncation fix, 15+ false positive cleanups, and why mapped is not detected.
Walk through writing a custom detector for `wget | sh` patterns. File location, the trait, registration, unit test, validation via make replay-qa.
Control-plane observability is mature; node-level eBPF detection is gap-y. Inner Warden as a DaemonSet, container escape detectors, mesh broadcast for fleet-wide blocking.
Real-world walkthrough of what attackers do in the first minute. Each step mapped to MITRE ATT&CK and what eBPF + behavioral detection sees vs what log-only tools miss.
How behavioral DNA identifies campaigns across IPs using SHA-256 hashing of attack patterns and union-find clustering. 47 IPs, 8 countries, one botnet.