Skip to content
Cisent Message us on WhatsAppMessage us

Manthys · decisions

Reflex

Our decision model. Reflex C 2 reads every log and decides what deserves the judge's attention; it doesn't write: it answers with a decision and its probability.

How it decides

One question. One probability.

Reflex C 2 reads a window of firewall logs and answers yes or no, with its probability. What it says yes to, it escalates to the judge, which reasons it through. The examples are synthetic and illustrate how it decides; the test is the 48-hour one, below.

Is there a port scan?

One source walks 7 ports

  1. 23:42:30 netcfg: applying new configuration (61881 ms)
  2. 23:42:30 DROP TCP 203.0.113.45:30872 → 198.51.100.7:8080 SYN
  3. 23:42:32 DROP UDP 192.0.2.23:60159 → 239.255.255.250:1900
  4. 23:42:32 DROP TCP 203.0.113.45:19849 → 198.51.100.7:143 SYN
  5. 23:42:34 DROP TCP 203.0.113.45:30719 → 198.51.100.7:21 SYN
  6. 23:42:35 DROP TCP 203.0.113.45:58214 → 198.51.100.7:80 SYN
  7. 23:42:37 DROP TCP 203.0.113.45:39137 → 198.51.100.7:445 SYN
  8. 23:42:37 DROP TCP 203.0.113.45:24260 → 198.51.100.7:110 SYN
  9. 23:42:38 DROP UDP 192.0.2.23:5353 → 224.0.0.251:5353
  10. 23:42:40 DROP TCP 203.0.113.45:13923 → 198.51.100.7:25 SYN

probability

0.99

threshold 0.50

Yesescalates to the judge

3 ports: not a scan yet

  1. 08:35:03 DROP UDP 192.0.2.12:51152 → 224.0.0.252:5355
  2. 08:35:04 DROP TCP 203.0.113.77:3056 → 198.51.100.7:8443 SYN
  3. 08:35:06 Started lldpd.service.
  4. 08:35:08 DROP TCP 203.0.113.77:48484 → 198.51.100.7:443 SYN
  5. 08:35:08 initializing plugin instances
  6. 08:35:09 connected to controller
  7. 08:35:11 netcfg: applying new configuration (18172 ms)
  8. 08:35:12 DROP 2 192.0.2.12 → 224.0.0.22
  9. 08:35:14 DROP TCP 203.0.113.77:38174 → 198.51.100.7:80 SYN

probability

0.01

threshold 0.50

Nodoesn't escalate

Normal network traffic

  1. 09:14:16 firewall: rules reloaded (17644 rules)
  2. 09:14:16 DROP UDP 192.0.2.12:137 → 192.0.2.255:137
  3. 09:14:16 DROP UDP 192.0.2.88:33061 → 192.0.2.1:53
  4. 09:14:17 DROP UDP 192.0.2.88:46076 → 239.255.255.250:1900
  5. 09:14:17 DROP 2 192.0.2.23 → 224.0.0.22
  6. 09:14:18 DROP UDP 192.0.2.88:137 → 192.0.2.255:137
  7. 09:14:20 DROP UDP 192.0.2.88:61651 → 224.0.0.252:5355
  8. 09:14:20 Received signal 1, reopening logfiles
  9. 09:14:21 DROP TCP 203.0.113.9:3259 → 198.51.100.7:443 SYN

probability

0.01

threshold 0.50

Nodoesn't escalate

Is there a brute-force attack?

7 attempts in a row on port 22

  1. 20:25:37 DROP TCP 198.51.100.23:7401 → 198.51.100.7:22 SYN
  2. 20:25:39 unifi-core.service: Succeeded.
  3. 20:25:41 DROP UDP 192.0.2.88:137 → 192.0.2.255:137
  4. 20:25:43 DROP TCP 198.51.100.23:4066 → 198.51.100.7:22 SYN
  5. 20:25:44 DROP TCP 198.51.100.23:24630 → 198.51.100.7:22 SYN
  6. 20:25:46 DROP 2 192.0.2.57 → 224.0.0.22
  7. 20:25:46 DROP TCP 198.51.100.23:35186 → 198.51.100.7:22 SYN
  8. 20:25:47 DROP TCP 198.51.100.23:60977 → 198.51.100.7:22 SYN
  9. 20:25:48 DROP TCP 198.51.100.23:31143 → 198.51.100.7:22 SYN
  10. 20:25:48 DROP TCP 198.51.100.23:51920 → 198.51.100.7:22 SYN

probability

0.99

threshold 0.50

Yesescalates to the judge

4 attempts on 3389: below the threshold

  1. 15:53:17 FW UDP 192.0.2.57:5353 → 224.0.0.251:5353
  2. 15:53:17 Configuration reload request received, reloading configuration;
  3. 15:53:19 DROP TCP 198.51.100.66:15225 → 198.51.100.7:3389 SYN
  4. 15:53:20 connected to controller
  5. 15:53:20 DROP TCP 198.51.100.66:18931 → 198.51.100.7:3389 SYN
  6. 15:53:22 DROP TCP 198.51.100.66:44858 → 198.51.100.7:3389 SYN
  7. 15:53:23 DROP TCP 198.51.100.66:51880 → 198.51.100.7:3389 SYN
  8. 15:53:25 connected to controller
  9. 15:53:27 FW UDP 192.0.2.23:44602 → 239.255.255.250:1900

probability

0.53

threshold 0.50

Yesescalates to the judge

Five sources, once each

  1. 12:31:29 DROP TCP 203.0.113.56:29211 → 198.51.100.7:22 SYN
  2. 12:31:31 DROP TCP 203.0.113.34:57884 → 198.51.100.7:22 SYN
  3. 12:31:31 DROP UDP 192.0.2.23:137 → 192.0.2.255:137
  4. 12:31:32 DROP TCP 203.0.113.90:22033 → 198.51.100.7:22 SYN
  5. 12:31:34 DROP TCP 203.0.113.78:20918 → 198.51.100.7:22 SYN
  6. 12:31:36 DROP UDP 192.0.2.88:5353 → 224.0.0.251:5353
  7. 12:31:36 initializing plugin instances
  8. 12:31:38 DROP TCP 203.0.113.12:3535 → 198.51.100.7:22 SYN
  9. 12:31:38 DROP UDP 192.0.2.57:58860 → 239.255.255.250:1900

probability

0.94

threshold 0.50

Yesescalates to the judge

Should the judge be alerted?

A scan forwarded to an inside host

  1. 08:25:58 statusMonitor: cpu 80% mem 83%
  2. 08:25:58 Started ntp.service.
  3. 08:26:00 systemd-tmpfiles-clean.service: Succeeded.
  4. 08:26:00 ALLOW TCP 203.0.113.200:40654 → 192.0.2.50:22 SYN
  5. 08:26:02 ALLOW TCP 203.0.113.200:15414 → 192.0.2.50:5432 SYN
  6. 08:26:02 ALLOW TCP 203.0.113.200:23443 → 192.0.2.50:3306 SYN
  7. 08:26:03 ALLOW TCP 203.0.113.200:8904 → 192.0.2.50:443 SYN
  8. 08:26:04 ALLOW TCP 203.0.113.200:59779 → 192.0.2.50:80 SYN
  9. 08:26:04 FW UDP 192.0.2.57:49072 → 192.0.2.1:123

probability

0.99

threshold 0.50

Yesescalates to the judge

Baseline only

  1. 15:06:42 DROP UDP 192.0.2.88:57090 → 224.0.0.252:5355
  2. 15:06:43 health: wan check ok (90136 ms)
  3. 15:06:45 mem avail: 39256 of 3900 MiB (41 %), swap free: 0 of 0 MiB
  4. 15:06:47 Reached target Network is Online.
  5. 15:06:48 statusMonitor: cpu 61% mem 48%
  6. 15:06:48 DROP UDP 192.0.2.23:5353 → 224.0.0.251:5353
  7. 15:06:49 DROP UDP 192.0.2.23:53546 → 224.0.0.252:5355
  8. 15:06:50 DROP 2 192.0.2.57 → 224.0.0.22
  9. 15:06:51 reading /etc/resolv.conf

probability

0.01

threshold 0.50

Nodoesn't escalate

Probability Manthys Reflex C 2 gives for each window, measured on October 12, 2026. The logs are synthetic.

48 hours, measured blind

27of 28

attacks detected in 48 hours of real traffic with injected attacks: port scans and brute force.

Measured blind: attacks were injected into 48 hours of real traffic from a lab firewall (7,401 windows). 28 were detectable: the reference rule sees them in some window. False alarms per day: 3 on “port scan”, 5 on “alert the judge” and 0 on brute force.

Reflex filters. Sentinel judges. Husker decides.

Where it sits

In front of the judge.

Manthys Reflex C 2

It watches every window.

It reads the logs and decides whether there's something to look at.

Manthys Sentinel Ultra 2

It reasons through what's doubtful.

What Reflex C 2 flags goes to the judge, which reasons it through.

Husker

It decides and records.

It acts and keeps a record of everything.

Two models

C for cybersecurity. G for general purposes.

Manthys Reflex C 2 · cybersecurity

It decides on security logs.

It reads every window of logs and decides, with its probability, whether there's something the judge needs to look at.

Manthys Reflex G · general purpose

In development.

It will bring Reflex's decisions to general-purpose work.

Every figure

Reflex C 2's table.

MeasureManthys Reflex C 2
Detection of injected attacks, of the detectable ones (port scans and brute force)27 of 28 [82–99%]
False alarms per day · port scan3 [1.1–6.5]
False alarms per day · alert the judge5 [2.4–9.2]
False alarms per day · brute force0 [0–1.8]
Decoys that fooled it0 of 8

Measured blind: the attacks were injected into 48 hours of real traffic from a lab firewall (7,401 windows), and the truth is marked by a reference rule; detectable ones are those the rule sees in some window. In brackets, the 95% confidence interval. The 10 false alarms on “alert the judge” come from a single 4-hour episode. It is a single network and a small sample (28 detectable attacks), and the reference rule is exact by construction.

Manthys

The other families.

Manthys Mind · generalist

It says “I don't know” when it doesn't know.

Explore Mind

Manthys Sentinel · cybersecurity

It reads the attack like an analyst.

Explore Sentinel

Manthys Nova · the largest scale

It flags its doubt. So you can verify.

Explore Nova

Book a meeting with the Cisent team.

Tell us what your organization needs to protect. We'll reply on WhatsApp and set up the meeting.

Message us on WhatsApp and let's book a meeting

or write to info@cisent.ai