Skip to content
NewCybercrime Report 2026 · Eight Chapters · A Post Is Not An Incident → Explore The ChaptersCybercrime Report 2026 → Explore

Odin Research / What is / Techniques and vulnerabilities

What Is JARM? TLS Server Fingerprinting for Threat Hunting

A simple way to group servers by how they answer TLS handshakes. It finds command-and-control fleets fast, and it collides with ordinary servers just as fast.

By
Odin by Brandefense
Updated
21 Nov 2026
Reading time
6 min read
Type
Technique

In short

JARM is an active TLS server fingerprinting method released by Salesforce. It sends ten crafted TLS Client Hello messages to a server and hashes the replies into a 62-character fingerprint. Servers with the same TLS setup share the same JARM, which helps hunters group command-and-control servers. Legitimate servers often share those fingerprints too.

Key takeaways

What it is
An open-source, active TLS fingerprinting tool from Salesforce. It describes how a server's TLS stack answers, not which program runs behind it.
How it works
Ten Client Hellos with different versions, ciphers and extensions. The first 30 characters record the choices, the last 32 are a truncated hash of the extensions.
Why it matters
C2 fleets built from one template often share a JARM, so a single value can surface hundreds of servers for review.
What Odin Research saw
One JARM covered 506 of 539 Brute Ratel C4 records in Monograph 02, yet the same value sat on VPN exits and a vendor lab. Collisions appear in almost every Odin study.
What to do first
Use JARM to find and group servers, then confirm with a certificate subject, a protocol check or endpoint data before you block anything.

What is JARM?

JARM is an active TLS server fingerprinting method created at Salesforce. It sends a server ten specially built TLS Client Hello messages, records how the server answers each one and turns the answers into a 62-character fingerprint. Servers with the same TLS configuration produce the same JARM, so hunters use it to group servers.

Salesforce engineers John Althouse, Andrew Smart, RJ Nunnally and Mike Brady introduced JARM on the Salesforce Engineering blog. The tool is open source on GitHub, written in Python and released under the BSD-3-Clause license.

JARM is popular in threat hunting because it is cheap and quick. You can scan a list of addresses, compare the results and spot servers that were set up the same way. It is a step up from banner grabbing, because a JARM reflects the TLS library and settings, which operators often leave at their defaults.

How JARM works

According to Salesforce, JARM works in three steps:

  1. Probe. It sends 10 TLS Client Hello packets. Each one varies the TLS version, the cipher list, the extensions and their order.
  2. Record. For each probe, it captures what the server chose in its Server Hello reply, or that it refused.
  3. Hash. It builds the 62-character fingerprint.

The fingerprint has two parts:

  • The first 30 characters encode the cipher and TLS version the server chose for each of the 10 probes. The value 000 means the server refused that probe.
  • The last 32 characters are a truncated SHA-256 hash of the server's extensions, leaving out certificate data.

The README notes a useful reading trick. Two fingerprints with the same first 30 characters but different last 32 suggest similar configurations that differ in their extensions. Salesforce also points out that JARM is a fingerprinting function, not a cryptographic one, built to be short enough for people to read and share.

JARM is active: you have to send probes to the server. Its passive counterpart, JA3/S, fingerprints TLS from traffic you already see.

How JARM is used to find C2 servers

Many command-and-control frameworks ship with a fixed TLS setup. Operators deploy them from templates, so whole fleets end up with one JARM.

Salesforce gave two examples when it introduced JARM. About 80% of live addresses on an abuse.ch Trickbot C2 list shared one fingerprint, which did not overlap with the Alexa Top 1 Million sites. A scan of port 443 by SecurityTrails likely found most Cobalt Strike C2 servers listening at the time.

Salesforce also listed other uses: checking that a fleet of your own servers shares one TLS configuration, grouping servers by configuration and spotting default applications. It suggested combining JARM with historical data, because malicious servers tend to be short-lived while legitimate ones stay put.

Why JARM alone is not enough

JARM measures the TLS stack, not the program. Salesforce itself noted false positives: Burp Collaborator, a point-of-sale system and some legitimate servers matched the Cobalt Strike fingerprint. It also said that linking fingerprints to specific software versions had not been studied exhaustively.

Odin Research's monographs list four common reasons for collisions:

  • Reverse proxies. Nginx, Caddy and Traefik in front of a C2 server replace its fingerprint with a generic one.
  • Shared libraries. Many Go programs use the same default TLS stack, so unrelated Go services look alike.
  • Cloud load balancers. Managed load balancers share one fingerprint across thousands of customers.
  • TLS 1.3. A TLS 1.3-only server gives a fingerprint full of refusals and hides its certificate from passive sensors.

That is why a JARM match belongs in the "lead" column until a second signal backs it up.

What Odin Research observed

JARM appears in almost every Odin Research monograph from 02 to 29. Some examples:

  • One value, many owners. In Monograph 02, one JARM covered 506 of 539 feed-labeled Brute Ratel C4 records. The same value also appeared on 90 records in a security vendor's demo lab, 4 ARTEX records and 3 TacticalRMM records.
  • VPN exits. Monograph 19 found 507 Brute Ratel C4 hits that were commercial VPN exits sharing that stock fingerprint. Blocking them would have cut off remote workers.
  • Template fleets. Monograph 02 also found one JARM on 233 of 276 VShell records, and one stack shared by the DarkSword and SuperShell feed labels.
  • Load balancers. Monograph 13 found 189 AWS load balancers with the exact JARM of AWS-hosted Interactsh servers. Only a certificate issuer and an application-layer check told them apart.
  • Useful clusters. Monograph 21 saw two SoftEther bridges share a JARM prefix. Monograph 26 found an Orcus RAT candidate and a XenoRAT candidate on one fingerprint, and Monograph 29 found 123 ARL nodes sharing one.
  • TLS 1.3 blind spot. In Monograph 02, all 16 Sliver operator listeners on port 31337 that answered returned a TLS 1.3-only JARM.

Across these studies, Odin's rule is the same: a JARM match can open an investigation, but it never closes one.

How to use JARM safely

  1. Hunt, then confirm. Use JARM to find candidates in your attack surface data or scan results. Confirm each one with a certificate subject or issuer, a protocol check or endpoint evidence.
  2. Compare prefixes. Matching first 30 characters with different endings often means the same TLS library with different extensions.
  3. Add history. A fingerprint that appears and disappears with a short-lived server is more interesting than one on a long-standing site.
  4. Never auto-block. Do not feed JARM-only matches into firewall blocklists. Treat them like any other false-positive-prone signal.
  5. Cover TLS 1.3. Pair JARM with passive server fingerprints and endpoint process data where certificates are hidden.

Monographs 02 and 19 publish JARM values for hunting, each meant to be paired with a certificate check. Their Suricata rules key on certificate subjects and protocol content rather than JARM. All rules are in the Defense Kit.

JARM compared with other TLS fingerprints

MethodActive or passiveWhat it fingerprintsBest use
JARMActive (sends 10 probes)How a server answers varied Client HellosGrouping internet servers before you have traffic
JA3/SPassiveClient and server hellos seen in your own trafficSpotting known tools in network logs
Certificate subjectEitherThe name a server presents in its certificateConfirming a family when TLS 1.2 exposes it

Frequently asked questions about JARM

What is JARM?

JARM is an active TLS server fingerprinting method from Salesforce. It sends ten crafted Client Hello messages to a server and turns the replies into a 62-character fingerprint that groups servers with the same TLS setup.

What does a JARM hash mean?

The first 30 characters record the cipher and TLS version the server picked for each of the ten probes, with 000 for a refusal. The last 32 characters are a truncated SHA-256 hash of the server's extensions.

Who created JARM?

Salesforce engineers John Althouse, Andrew Smart, RJ Nunnally and Mike Brady introduced it on the Salesforce Engineering blog. The tool is open source under the BSD-3-Clause license.

JARM vs JA3: what is the difference?

JARM is active: it probes a server and fingerprints its answers. JA3 and JA3S are passive: they fingerprint client and server hellos in traffic you already capture.

Can I block traffic based on JARM?

Not on its own. Reverse proxies, shared Go TLS stacks, cloud load balancers and VPN exits produce the same fingerprints as C2 servers. Confirm with a second signal first.

Is JARM still useful?

Yes, for hunting and grouping. Odin Research uses it in almost every study to cluster servers, then confirms each cluster with certificates, protocol checks or endpoint data.

Does JARM work with TLS 1.3?

It still produces a fingerprint, but TLS 1.3-only servers refuse many probes, and TLS 1.3 hides the certificate from passive sensors. Pair it with passive fingerprints and endpoint data.

How do I get a JARM fingerprint?

Run the open-source JARM tool from Salesforce's GitHub repository against a host and port, or use an internet scanning service that records JARM values.

Sources

  1. Salesforce Engineering, "Easily Identify Malicious Servers on the Internet with JARM". engineering.salesforce.com/easily-identify-malicious-servers-on-the-internet-with-jarm-e095edac525a/ (checked 21 November 2026)
  2. Salesforce, JARM repository and README, GitHub. github.com/salesforce/jarm (checked 21 November 2026)

Related tools and terms

All Odin Research explanations · Brandefense glossary

Odin Research

Get the rules, not just the definition

Every Odin Research report ships with verified indicators and detection rules you can load today.

Open the Defense Kit