Deception Check

FIRST-PARTY RESEARCH / SSH ACCESS

The SSH key they
tried to leave behind.

A password reset does not remove an unauthorized SSH key. Follow the attempted change, then check whether the server could accept it.

A login is one way in. An added SSH key can become the next.

The same complete public key appeared in key-write commands across 3,639 sessions on our research decoys. Matching content hashes appear in virtual-file capture records for 3,638 of those sessions. The useful lesson is the change in trust: investigate what access an intruder tried to leave behind, rather than stopping at the login or the password reset.

Start with the access the server would accept

An authorized public key lets its matching private key authenticate to an account when the server's policy permits it. Adding an unauthorized entry can create a way back in. MITRE ATT&CK maps this behavior to SSH Authorized Keys, T1098.004.

On one host, connect three records: the effective key source, the change that added the key, and any later authentication. Do not assume every service reads the default ~/.ssh/authorized_keys file; an alternative file or AuthorizedKeysCommand can supply it.

Three questions for an SSH-key investigation
QuestionEvidence to requestDecision it supports
Where does this account get its keys?Effective SSH configuration for the host, account, and connection context; key source and restrictions.Whether the changed source could grant usable access.
Who changed it, and was that approved?Before/after key content, file or provider audit event, writing process, session, and change approval.Expected administration or an unexplained trust change.
Was that key accepted later?SSH authentication records, including a key fingerprint when collected; host, account, time, and source.Observed use, or a specific collection gap to resolve.
A missing fingerprint field is a logging gap, not evidence that the key was never used.

One key. Two kinds of retained evidence.

3,639

Distinct sessions submitted the same complete SSH public key in a key-write command.

COMMAND RECORD / OCT 1 TO 9, 2026 UTC
3,638

Of those sessions also matched the content hash in a virtual-file capture record.

MATCHED IN THE SAME DECOY AND SESSION
These are research-decoy records. They establish a submitted key-write operation and a matching capture record, not later successful access to a production host. One submission session had no matching capture in the selected export.

The command record tells us what was requested. The virtual-file capture record gives us a second check: its content hash matches the submitted key line. Joining them within the same session is stronger than finding a familiar filename somewhere in a large command log.

  1. 01 / REQUESTED CHANGESubmit a key-write command

    The complete key is present in the recorded input.

    OBSERVED / 3,639 SESSIONS
  2. 02 / RETAINED CONTENTMatch the virtual-file capture

    The captured content hash joins to the key submission in the same session.

    OBSERVED / 3,638 MATCHED SESSIONS
  3. 03 / PRODUCTION FOLLOW-UPCheck the accepting SSH service

    Review effective configuration, approval, and authentication records to establish usable access and later use.

    PRODUCTION CHECK / REQUIRES HOST RECORDS
The observed evidence leads into a separate check on the production host.

ONE RETAINED SESSION / OCTOBER 1, 2026 UTC

Inspect the two records.

Step through one actual sequence. Presentation timing is slowed for reading; the recorded gap was about 180 milliseconds.

Both records remain readable below.

01:03:28.433307 UTC / SUBMITTED INPUT

Write a key to the account's SSH key file

The retained command contains the complete public key. Raw command, key bytes, and infrastructure identifiers are omitted from this shared view.

01:03:28.613805 UTC / VIRTUAL-FILE CAPTURE

Retain matching key content in the same session

The capture record carries a hash matching the submitted key line plus its newline.

Retained timestamps, joined within the same decoy and session. Source addresses and original session identifiers remain private.

The comment is a label. Fingerprint the key.

The observed key matches the complete public key published in Kaspersky's April 2025 Outlaw investigation. We compared the decoded key material using its SHA-256 fingerprint:

SHA256:MkYY9qiVsFGBC5WkjoClCkwEFW5iSjcGQF7m4n4H7Cw

Use that as an artifact pivot, not an actor verdict. The same public key can be copied by different operators. The trailing comment can also change without changing the key; OpenSSH does not use that comment for authentication.

On a preserved copy of an authorized-key file, ssh-keygen can list key fingerprints. This command reads a local file; it does not connect to a server or install a key.

ssh-keygen -l -E sha256 -f ./authorized_keys.evidence

Keep the original authorized-key line and its restrictions alongside the fingerprint. A matching key with different restrictions can have different access implications. Preserve host-by-host timelines when searching for the same fingerprint elsewhere.

Turn the finding into an owned decision

  1. Preserve the change.Retain original key content, relevant audit events, authentication logs, and clock context before rotation or retention removes them.
  2. Confirm what it could authorize.Check the service's effective key source, account policy, and key restrictions. Compare the change with approved administration.
  3. Look for use.Join subsequent authentication to the host, account, time, and key fingerprint where available. Name missing telemetry explicitly.
  4. Recover access with the owner.Coordinate removal or revocation of unauthorized keys, account recovery, and a test that approved operational access still works. A password reset alone does not remove a separate key entry. A usable return route depends on the server still accepting that key.

Deception Check gives us a controlled place to retain an attempted key installation. The production response still needs the records that explain whether a particular host would trust that key and whether it did.

TAKE IT TO ONE HOST

Can you join the change to its next use?

The worksheet records the effective key source, fingerprint, change approval, later authentication, and the owner of the next check.

Download the two-page worksheetUse the editable worksheet

If a new SSH key appeared tonight, which record would your team check next?

Methods and scope

We counted distinct decoy-and-session pairs in retained SSH input records from October 1 through October 9, 2026 UTC. We extracted complete RSA public-key material from commands targeting authorized-key files, decoded it, and computed the SHA-256 fingerprint. A second join looked for virtual-file capture events carrying the hash of the complete observed key line plus its newline in the same decoy and session.

The counts describe sessions, not distinct people or production compromises. The selected export does not establish the cohort's original authentication method or later acceptance of the key. This review checked the sensor-recorded content hash, rather than rehashing separate artifact bytes. The methods distinguish public-key identity from the full captured file content; changing a comment can change a file hash while leaving the public-key fingerprint unchanged.

Sources and technical references