Only alerting on the firewall threats that got through
This week EdTech Magazine wrote up the State of K-12 Cybersecurity 2026 report from CIS and MS-ISAC. About half of schools had a cyber incident last year, and the line that stuck with me was about staffing: two-thirds of districts have a cybersecurity team of one or two people, and about one in ten have no dedicated team at all.
That's the reality I build alerting for. A small team can collect every log in the building, but it can't read them all. So when I connected our firewall's Threat logs to Wazuh this week, the real question wasn't "how do I get the logs in?" It was "which of these should actually interrupt someone?"
The problem with firewall threat logs
A Palo Alto firewall writes a Threat log every time traffic matches an exploit, malware or spyware signature. Most of those are good news: the firewall saw something bad and dropped it. Internet-facing services get poked at all day, and a blocked attempt needs nothing from me. If every Threat log became an alert, the alerts would be ignored within a week.
Once the logs were flowing, I looked at what had arrived over the first couple of days. Not one would have needed a person: the critical ones were all blocked, and the rest were low-level scanner noise. That told me what the rules should look for. Two kinds of Threat log do need someone:
- A high or critical threat the firewall only logged. Depending on how a profile is set, the firewall can match a signature and still let the traffic through (an action of
alertorallow). Someone needs to check the target. - A sinkholed DNS lookup. A device asked for a known-bad domain, and the firewall answered with a sinkhole address instead. The lookup was stopped, but the device that asked may be infected.
Everything else stays in the Wazuh dashboard, searchable but quiet.
What you need
- A Wazuh manager receiving Palo Alto Threat logs over syslog (the built-in Palo Alto decoder and rules handle the parsing)
- Something that acts on a high alert level: an integration, email alerts or a dashboard filter
- A way to test rules before you restart the manager (
wazuh-logtestships with Wazuh)
Rule 1: a threat that wasn't blocked
Wazuh already has built-in rules for Palo Alto Threat logs by severity. My rule sits on top of the high and critical ones and fires only when the action is not a block:
<rule id="100200" level="12">
<if_sid>64503,64504</if_sid>
<field name="type">^THREAT$</field>
<action type="pcre2" negate="yes">(?i)^(?:drop|reset-both|block-url|deny|sinkhole|...)$</action>
<description>Palo Alto: $(severity) threat was not blocked (action $(action)): $(threat_id)</description>
</rule>
The detail I like most is the negate="yes". I could have written "alert when the action is alert or allow", but then any action I forgot, or one a future firewall update adds, would fall through silently. Instead the rule lists every action that stops traffic and alerts on anything else. If I'm wrong, I get an extra alert, not a missed one. (The shortened list above is just for reading; the full one is in the rules file linked below.)
Level 12 matters because that's the level our Wazuh integration forwards to n8n. From there it posts a card to the security team's chat space and opens a case in the small security app we use to track what needs a look.
Rule 2: a sinkholed lookup
The sinkhole rule is simpler: any severity, action sinkhole. It hangs off all four severity rules so it doesn't depend on which built-in rule happens to match first.
There's one caveat I wrote down before it ever fired. If devices look names up through an internal DNS server, the sinkholed lookup that crosses the firewall may come from that DNS server, not the device. The alert would then really say "some device asked", and finding which one means checking the DNS server's query log. I'll know which way ours goes on the first real alert.
One problem, one alert
One scan or one infected laptop can produce the same Threat log dozens of times. Each level 12 rule has a partner at level 3 that takes over for repeats:
<rule id="100201" level="3" frequency="2" timeframe="3600">
<if_matched_sid>100200</if_matched_sid>
<same_field>threat_id</same_field>
<same_field>destination_address</same_field>
<description>Palo Alto: repeat of an unblocked $(severity) threat within an hour</description>
</rule>
For unblocked threats, repeats are grouped by threat and destination for an hour, because the destination is what someone has to check. For sinkholes, they're grouped by signature and source for 24 hours, because the source is the possibly infected device. A different threat or a different device still alerts on its own, and the repeats stay searchable at level 3.
Testing before it goes live
I don't restart a Wazuh manager on hope. The change script has a preview mode that copies the rules in, runs the config test, feeds a set of sample log lines through wazuh-logtest and checks that each one lands on the rule I expect, then removes the file again. It also sends the same line three times to make sure the first one alerts and the next two drop to the repeat rule. Only then does apply mode back up the config, install the file, run the same tests again and restart. If any test fails, the file comes back out and nothing restarts.
The sample lines are made up from a real line's format, with documentation addresses swapped in, because no real unblocked or sinkholed line had arrived yet. I also only tested the 24-hour repeat window with lines seconds apart, so that part has only been proven at short range.
Where it stands
The rules went live on November 20. As I write this, neither one has fired on real traffic, which is what I'd expect from that first look at the logs. That's fine: the point is that when one does fire, it's worth someone's time. If the sinkhole rule turns out noisy, the plan is to drop it to a lower level or narrow it by severity.
One last rule I follow: Threat logs can include a staff member's sign-in name and device address, so those details stay in Wazuh and the case app. They don't get copied into tickets, chat threads or notes.
I put a sanitized copy of the rules, with install and rollback steps, in my CtrlAltUpgrade repo on GitHub. Check the parent rule IDs against your Wazuh version before you use it.
The takeaway for a one- or two-person team: don't alert on what your tools stopped. Alert on what got through, and on the signs that something inside might already be wrong. Keep the rest where you can search it when you need it.