Before sshhole was disclosed, 104.244.79.113 was already sending ordinary SSH password guesses. On 6 November, it submitted passwords consisting of exactly 257 A characters against root: the minimum length that can trigger the newly disclosed authentication bypass. In the post-disclosure SSH traffic we reviewed from this address, those were the only passwords it sent.
Three other addresses previously seen guessing SSH passwords also began submitting oversized passwords. A fifth, previously seen probing web services, appeared in the SSH set. These sources provide the clearest evidence of adoption: a new vulnerability entering activity already in progress.
OpenSSH released a workaround, version 10.6p2, on 5 November. OHIIHO observed matching attempts that day, and 24 source addresses across 5–7 November. This note examines their payloads, account choices and SSH fingerprints. The counts describe our observations, not Internet-wide prevalence.
Previdian had already reported exploitation by 6 November. Previdian founder Ryan Dewhurst identified 89.116.89.18 in a statement to Help Net Security; that address also appears in our dataset. This analysis complements that reporting with additional observations and behavioural detail.
Existing sources, a new authentication attempt
The four addresses with earlier SSH password-guessing activity did not all adopt the same implementation.
| Source address | Post-disclosure submissions | Interpretation |
|---|---|---|
104.244.79.113 | Exactly 257 As against root | A switch to the minimum triggering length in the reviewed traffic |
77.239.124.175 | 300 As against several account names | The bypass added to an existing password-guessing source |
213.177.179.80 | Random-looking 264-byte strings against root and alpine | A different payload format, continuing into 7 November |
122.160.224.130 | Repeated As at 300 and 309 bytes against root and alpine | The same vulnerability approached with another client profile |
The source previously seen probing web services was 212.224.107.115; it submitted 300 As against alpine.
This is continuity at the source-address level. It does not establish that the same person controlled an address before and after disclosure. Nevertheless, the change in submitted passwords is concrete: addresses already present in the traffic began sending values aligned with a newly published bypass.
For defenders, a familiar SSH source can change what it is testing without changing its address. A history of failed password guesses is insufficient context for the next authentication event.
Most attempts converge on a public example
The first matching submissions in the reviewed data arrived on 5 November, the day the fix and public disclosure appeared. No matching submissions were found in the pre-disclosure SSH data reviewed for this analysis.
| UTC date | Source addresses first observed submitting matching passwords |
|---|---|
| 5 November | 7 |
| 6 November | 15 |
| 7 November | 2 |
These are newly observed addresses within this dataset, not newly created infrastructure or counts of operators. Each address is counted once, on its first qualifying day.
Most sources sat in commercial hosting and cloud ranges; four exited through a commercial VPN and three came from consumer broadband. None was a Tor exit.
Across the qualifying attempts, 57% submitted exactly 300 As. Public reproductions, including the reporter’s repository and predkambrij/sshhole-repro, use that string. Its prevalence is consistent with reuse of public examples.
Other submissions included repeated characters at different lengths and random-looking strings. All observed passwords in this set were ASCII and between 257 and 805 bytes, within the reported 257–1,024-byte triggering range. Length alone does not establish intent; here, the timing, repeated fillers and account choices support treating the submissions as suspected sshhole attempts.
root and alpine accounted for 81% of qualifying attempts. Other targets included admin, deploy, git, ansible, docker, ubuntu, debian, user and pi. Two addresses also tried keyonly, a name suggestive of testing for an account intended to use SSH keys.
Five addresses, one tooling profile
The 24 source addresses used nine distinct HASSH values. HASSH fingerprints the ordered algorithm lists advertised during SSH key exchange; it describes a negotiation profile, not a unique tool or operator. Client configuration can change it, and different implementations can share it. The HASSH specification explains the construction.
One group combined four addresses identified as Swedish commercial VPN exits—185.209.199.50, 185.209.199.72, 185.213.154.234 and 185.213.154.248—with Spanish broadband address 79.116.138.30. All five advertised SSH-2.0-paramiko_5.0.0, used HASSH 6372ee6957562199b2fb773b0be1bf34, and submitted 300 As. Their observed account choices were confined to root and alpine; one address tried only root.
That combination is a useful hunting pivot. It is compatible with a common script or configuration, including a public one used independently.
Another HASSH, a2de0f306611e0957be704f5b0e35a82, appeared under both paramiko_4.0.0 and OpenSSH_9.6 banners. The banner is a client-supplied string; the fingerprint adds negotiation detail. Neither is an authenticated identity. Correlate banner, fingerprint, password shape, account choices and source history rather than treating any one of them as attribution.
AI-assisted discovery, rapid adoption
There is a documented role for agents on the discovery side. In his technical write-up, Özgün Kültekin describes using agent swarms to test OpenSSH builds across libc, PAM and account configurations. Separately, the OpenSSH release notes say that bugs found with AI tools are later rediscovered independently by other researchers, so adversaries can likely find them too; the project will release fixes more often for now.
The traffic analysed here does not establish AI use by the sources submitting these passwords.
Exposure and investigation
sshhole, tracked as CVE-2026-123158, affects particular OpenSSH configurations on musl-based Linux. Without PAM, an oversized password can cause crypt() to return *; that failure value can match an account whose stored password field is also exactly *. The reported triggering range is 257–1,024 bytes. A usable SHA-256-crypt or SHA-512-crypt hash on another account supplies the salt needed for this path. The reporter’s repository documents the mechanism and checks.
For Alpine, verify the package and effective configuration against the distribution advisory. Exposure requires password authentication, the affected non-PAM path, an eligible * account and another account with a $5$ or $6$ hash. A * field means no usable password hash; it is not an empty password. OpenSSH’s default PermitRootLogin prohibit-password blocks password login as root; this route requires PermitRootLogin yes. Affected official Alpine tiny-cloud images also provide an alpine account with a * password field and passwordless doas. These images are not initially exploitable through this bypass, but adding a local $6$ password hash enables the vulnerable path described above, allowing access as alpine and escalation to root. Updated images start at Alpine 3.21.10, 3.22.8, 3.23.8 and 3.24.4.
Alpine lists the following fixed openssh package versions:
| Branch | Fixed version |
|---|---|
| edge | 10.6_p2-r0 |
| 3.24 | 10.3_p1-r2 |
| 3.23 | 10.2_p1-r1 |
| 3.22 | 10.0_p1-r11 |
| 3.21 | 9.9_p2-r1 |
| 3.20 | 9.7_p1-r6 |
| 3.19 | 9.6_p1-r3 |
| 3.18 | 9.3_p2-r4 |
| 3.17 | 9.1_p1-r7 |
| 3.16 | 9.0_p1-r6 |
Restart sshd after upgrading. If an update is unavailable, disable password authentication. Confirm working key access first and keep the current administrative session open while testing a new connection.
Configuration checks must account for Match blocks. Plain sshd -T does not evaluate every possible connection context; use sshd -T -C with the relevant user and connection parameters, as described in the OpenSSH manual.
For investigation, Wakefield recommends checking password-login events for accounts without usable passwords. Correlate Accepted password for events with accounts that had a * password field at the time of login. An event of that kind on an exposed host warrants incident investigation. Today’s account configuration alone cannot establish its historical state. Standard SSH authentication logs do not include the submitted password length.
The companion Sigma rule selects password logins to root and alpine as triage candidates. It cannot determine their password-field state or prove use of this bypass. Scope it to relevant hosts, correlate account history, and extend it to other eligible accounts. Legitimate password logins to these names are expected false positives.
A search-index blind spot
For teams whose authorised telemetry already includes submitted credential values, inspect the mapping before hunting. If the field used for exact-match searches has ignore_above: 256, longer values will not be indexed in that field: they can be absent from exact-match queries and aggregations while the documents remain present.
Where this limit applies, read the value from the stored document source, where available, or query the index’s record of ignored fields. Do not infer absence from an empty keyword aggregation. The limit counts characters; this case uses ASCII, so character and byte lengths coincide. Some columnar index modes handle over-limit values differently; check your engine’s documentation. For future detection, a numeric byte-length field avoids dependence on indexing the password itself.
Scope and downloads
These observations describe suspected exploitation attempts, not confirmed compromises of third parties. Shared payloads and fingerprints support tooling correlations. They do not establish operator identity, intent or the use of AI. Some sources may be exposure scanners or researchers.
The useful change is visible in the traffic: previously active SSH sources began submitting passwords tailored to the bypass on the disclosure timescale. Defenders should pair source history with exposure checks and authentication outcomes, especially for accounts intended to accept keys only.
- 📥 Source indicators, CSV: 24 addresses, ASN and country enrichment, day-level observations, banners, HASSH, password lengths and shapes, and accounts tried. The
seen_before_disclosurecolumn covers any earlier SSH or web activity. Enrichment reflects the supplied snapshot; geolocation is not operator location. - 📥 SSH password-login triage rule, Sigma: experimental; requires host and account context. The fingerprints are pivots, not standalone malicious signatures. Apply expiry and review to IP-based blocks, particularly for VPN exits and broadband addresses.