Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
AiSOC auditd profile
A drop-in audit.rules profile that gives the auditd
connector enough signal to drive the bundled Linux endpoint detection
rules (detections/endpoint/linux-*.yaml) without running a
host-side AiSOC agent.
The connector tails /var/log/audit/audit.log directly and matches on
the key= field every rule attaches via -k aisoc_*. The connector's
severity heuristic is a pure function of the key prefix:
| Key prefix | Severity |
|---|---|
aisoc_critical_* |
critical |
aisoc_priv_esc_* |
high |
aisoc_persistence_* |
medium |
aisoc_exec |
medium |
aisoc_watch_* |
low |
aisoc_audit_* |
low |
Highest-priority match wins, so an explicit aisoc_critical_* key beats the
generic aisoc_exec bucket. The table above is _KEY_SEVERITY_PREFIXES in
auditd.py; a
modification of a path in _HIGH_RISK_PATHS (/etc/passwd and friends) is
lifted to high regardless of the key.
This means the SOC analyst sees the same severity in the AiSOC console as the rule author intended at policy-write time — no second-guessing.
Install
Tested on Ubuntu 22.04, Debian 12, RHEL 9, Amazon Linux 2023.
# 1. Install auditd if it isn't already.
sudo apt-get install -y auditd # Debian / Ubuntu
sudo dnf install -y audit # RHEL / Fedora / Amazon Linux
# 2. Drop the profile into rules.d (NOT directly into audit.rules —
# augenrules will compose the final ruleset for you).
sudo install -m 0640 -o root -g root \
aisoc.rules /etc/audit/rules.d/99-aisoc.rules
# 3. Reload the kernel ruleset.
sudo augenrules --load
# 4. Confirm the rules are live.
sudo auditctl -l | grep aisoc_
You should see 47 rules with key=aisoc_* attached. If you see zero,
re-check that audit.rules.d isn't being clobbered by a CIS / STIG
benchmark profile and that auditd itself is running
(systemctl status auditd).
Connect AiSOC
Add the connector instance from the AiSOC console (or the API):
| Field | Value (example) |
|---|---|
| Host label | web-prod-01.eu-west-1 |
| Audit log path | /var/log/audit/audit.log |
| Cursor path | (leave blank — defaults to <audit_log_path>.aisoc-cursor) |
The connector needs read on the audit log and read/write on the
cursor file. The cleanest fit is to add the AiSOC service account to
the local adm group (which already owns /var/log/audit/):
sudo usermod -aG adm aisoc
…and re-login the AiSOC service so the new group takes effect.
What gets detected, end-to-end
The profile emits 47 rules across 22 distinct aisoc_* keys. The Linux
endpoint detections that consume this telemetry live under
detections/endpoint/; the ones most directly tied to the keys in this
profile are:
audit.rules key prefix |
Related detections in detections/endpoint/ |
|---|---|
aisoc_critical_memfd |
linux-memfd-create-then-execve.yaml, linux-fileless-via-procfd.yaml |
aisoc_critical_exec_tmp |
linux-chmod-plus-x-tmp.yaml, linux-large-stage-tmp.yaml, linux-ld-preload-set-tmp.yaml |
aisoc_critical_identity_write |
linux-passwd-modified.yaml, linux-shadow-read.yaml |
aisoc_critical_sudoers_write |
linux-auditd-sudoers-tampering.yaml, linux-sudoers-d-add.yaml, linux-nopasswd-sudo-line.yaml |
aisoc_critical_pam_write |
linux-pam-d-modified.yaml |
aisoc_critical_ssh_config |
linux-auditd-ssh-config-tampering.yaml, linux-ssh-config-permitrootlogin-yes.yaml |
aisoc_critical_authorized_keys |
linux-authorized-keys-bulk-append.yaml, fim-ssh-authorized-keys-changed.yaml |
aisoc_persistence_cron |
linux-cron-d-write.yaml, linux-anacron-job-add.yaml, linux-at-job-create.yaml |
aisoc_persistence_systemd |
linux-auditd-systemd-persistence.yaml, linux-systemd-timer-create.yaml |
aisoc_persistence_initd |
linux-initd-script-add.yaml |
aisoc_priv_esc_module_load |
linux-auditd-kernel-module-load.yaml, linux-kernel-module-load-insmod.yaml |
aisoc_audit_self_tamper |
linux-auditctl-disable.yaml, linux-auditd-stopped.yaml |
aisoc_watch_nss |
linux-nss-module-installed.yaml |
This is a topical index, not a wiring contract: a detection fires on the
normalised event, not on the audit key, so enabling a key does not guarantee
a specific rule matches on your hosts. docs/detections/truth-table.md is
the generated record of which rules the engine actually loads.
If you author a new rule against this profile, follow the same convention — pick a key with a documented prefix, and the rest of the pipeline (severity, console grouping, eval grading) lights up for free.
Tuning
The profile is intentionally conservative. Two knobs you'll want to consider on real hosts:
- Backlog limits.
-b 8192is enough for a quiet web tier, way too small for a busy database. Bump it to 32768 if you seeaudit_lost > 0in/var/log/audit/audit.log. - Lockdown. The trailing
-e 2is commented out so you can iterate. Once stable, uncomment it; an attacker can no longerauditctl -e 0you without a reboot, and the reboot itself becomes a high-confidence signal.
Why no host-agent?
This profile + the file-tail connector exists because a host-agent is a 5–7 day Go project we deferred to a later release. The trade-off:
- ✅ Zero net-new code on the customer host. Stock
auditdonly. - ✅ Works on any distro with
auditd(RHEL family, Debian family, Amazon Linux, SUSE). - ✅ The connector is the only AiSOC-specific surface, and it lives on the AiSOC side.
- ⚠️ Requires the AiSOC service to read
/var/log/audit/audit.log, which means either group membership inadmor a sidecar shipper (rsyslog/Vector/Fluent Bit forwarding the file). - ⚠️ Sub-second latency depends on poll interval; default is 5 minutes.
Drop the per-instance
poll_interval_secondsto30for tier-1 hosts.
When the host-agent ships, this profile stays exactly the same — the agent will read the same log file, parse it with the same library the connector uses, and emit identical normalized events.