Mirage exploitation snapshot – November 2026
Mirage exploitation snapshot, November 2026: ten CVEs, all in KEV, and one bug taking 85% of attempts
Internet-facing services (web, VNC, RDP, databases, SSH, mail, VPN)
In the 24 hours to 14 November 2026, OffSeq's Mirage honeypot fleet saw 10,752 source addresses; 878 of them (8.2%) met the rules for the public block list. Over the week, the ten exploited CVEs in the feed were all in CISA's Known Exploited Vulnerabilities catalog – and 85% of all exploit attempts went to a single bug, Citrix Bleed 2, from just five addresses.
OffSeq
7 min read
Key facts
- Source
- Mirage Feed snapshot of 13 November 2026 (generated 14 November, 02:48 UTC)
- Window
- 7 days to 13 November; trend 22 October – 13 November
- Sources seen, last 24 h
- 10,752 IPs · 1,136 ASNs · 153 countries
- Listed attackers, 7 days
- 1,693 IPv4 · 376 ASNs · 68 countries
- Excluded
- 8,427 research scanners · 250 VPN / Tor exits
- Exploited CVEs in the feed, 7 days
- 10 (all in CISA KEV)
- Exploit attempts, 7 days
- 60,463
- Credential pairs tried, last 24 h
- 103,583
- Credit
- OffSeq – Mirage
§ 01. The week in numbers
The Mirage fleet runs decoy services across several networks and publishes what it sees as the Mirage Feed. In the 24 hours before the 13 November snapshot, the fleet logged 677,745 events from 10,752 source addresses in 1,136 networks and 153 countries.
An address only reaches the public block list if it completed a TCP handshake and sent application data, was seen by at least 3 sensors in 2 networks within 7 days, is not a known research scanner or allowlisted infrastructure, and was first seen at least 24 hours earlier. Of the 10,752, 878 qualified (8.2%). Over the 7-day window the list held 1,693 addresses.
| Measure | Value |
|---|---|
| Source addresses seen, last 24 h | 10,752 |
| Listed and active, last 24 h | 878 |
| Listed attackers, 7 days | 1,693 |
| Research scanners seen and excluded, 7 days | 8,427 |
| VPN and Tor exits seen attacking, 7 days (not blocked) | 250 |
| Credential pairs tried, last 24 h | 103,583 |
Research scanners outnumbered listed attackers five to one. They are published separately so they can be allowlisted rather than blocked.
§ 02. Three weeks of listed attackers
The feed keeps daily totals, without addresses, in history.json. Between late October and 13 November the 7-day list rose from about 1,700 addresses to 3,974 and fell back to 1,693.
| Snapshot day | Active, 24 h | Listed, 7 days | Added | Removed | Research scanners |
|---|---|---|---|---|---|
| 2026-10-24 | 789 | 1,688 | 137 | 155 | 5,412 |
| 2026-10-28 | 1,075 | 2,059 | 574 | 303 | 6,102 |
| 2026-10-30 | 1,936 | 3,232 | 1,591 | 418 | 6,721 |
| 2026-11-03 | 1,543 | 3,974 | 420 | 233 | 7,550 |
| 2026-11-07 | 860 | 3,105 | 296 | 522 | 10,146 |
| 2026-11-10 | 841 | 2,364 | 154 | 451 | 9,115 |
| 2026-11-13 | 878 | 1,693 | 199 | 290 | 8,427 |
history.json; snapshots for 27 and 29 October are missing.Between 28 and 30 October 1,591 addresses were added (the 29 October snapshot is missing), and the 7-day list peaked at 3,974 on 3 November. Entries expire 7 days after their last activity, so the list drained over the following week. The history does not store addresses, so it cannot show which campaign drove the rise. The feed's filters also changed on 29 October and 4 November (crawler allowlist, shared-address handling); day-to-day steps around those dates partly reflect the rules, not attacker behaviour.
§ 03. Ten CVEs, all in KEV
The feed lists ten CVEs exploited by its listed sources in the 7 days to 13 November, with 60,463 attempts in total (the fleet saw 12 in all; the other two did not pass the listing rules). All ten are in the CISA KEV catalog, and all ten sit above the 99.5th percentile of EPSS.
| CVE | Vulnerability | Attempts | Sources | In KEV since | EPSS percentile |
|---|---|---|---|---|---|
CVE-2025-5777 | Citrix Bleed 2 | 51,606 | 5 | 2025-07-10 | 99.96 |
CVE-2017-9841 | PHPUnit eval-stdin RCE | 6,495 | 48 | 2022-02-15 | 99.99 |
CVE-2021-41773 | Apache HTTP Server path traversal to RCE | 1,960 | 23 | 2021-11-03 | 99.98 |
CVE-2024-4577 | PHP-CGI argument injection | 237 | 59 | 2024-06-12 | 99.98 |
CVE-2026-109618 | PHP2Shell | 127 | 49 | 2026-10-26 | 99.99 |
CVE-2019-19781 | Citrix ADC path traversal | 22 | 14 | 2021-11-03 | 99.99 |
CVE-2025-3248 | Langflow unauthenticated RCE | 7 | 3 | 2025-05-05 | 99.98 |
CVE-2018-13379 | FortiOS SSL-VPN path traversal | 6 | 2 | 2021-11-03 | 99.99 |
CVE-2023-4966 | Citrix Bleed | 2 | 1 | 2023-10-18 | 99.99 |
CVE-2023-20198 | Cisco IOS XE Web UI privilege escalation | 1 | 1 | 2023-10-16 | 99.95 |
cves-7d.csv; KEV dates from the CISA catalog, EPSS percentile from the FIRST API (rounded down), both 14 November 2026.Volume and reach tell different stories. Citrix Bleed 2 took 85.4% of all attempts, yet came from only five addresses – a few hosts hammering one bug. PHPUnit (48 sources), PHP-CGI (59) and PHP2Shell (49) are the broad, many-source campaigns. Seven of the ten bugs have been in KEV since 2024 or earlier; the oldest, PHPUnit, was disclosed in 2017.
§ 04. PHP2Shell: seen before KEV
One CVE in the table reached the fleet before it reached KEV. PHP2Shell (CVE-2026-109618 and CVE-2026-109619) was disclosed on 21 October 2026. The first attempt on the fleet using the session upload-progress trigger both flaws depend on arrived at 16:21 UTC that day. CISA added both CVEs to KEV on 26 October, five calendar days later (KEV records the date, not the time).
The fleet was not first: Amazon's MadPot recorded exploitation attempts on 17 October, four days before disclosure. Three weeks on, PHP2Shell is still active, with 127 attempts from 49 addresses in the last 7 days.
§ 05. Where the traffic comes from
The 1,693 listed addresses span 68 countries and 376 networks. Location reflects where attackers rent or compromise machines, not where they are.
| Country | Listed addresses | Share |
|---|---|---|
| United States | 623 | 36.8% |
| Singapore | 166 | 9.8% |
| China | 128 | 7.6% |
| Netherlands | 96 | 5.7% |
| Germany | 78 | 4.6% |
| Network (ASN) | Listed addresses |
|---|---|
AS132203 Tencent | 166 |
AS14618 Amazon | 88 |
AS45102 Alibaba (US) | 68 |
AS8075 Microsoft | 59 |
AS14061 DigitalOcean | 58 |
AS401626 Netiface America | 55 |
AS396982 Google | 52 |
Eight cloud ASNs (Tencent, Amazon, both Alibaba ASNs, Microsoft, DigitalOcean, Google and Akamai/Linode) account for 550 addresses, 32.5% of the list. Blocking by cloud provider is not practical; blocking by address, with a short expiry, is.
§ 06. What it tries: credentials, paths and clients
Half the listed addresses (869, 51.3%) sent web exploits. VNC password guessing came next (364), then database logins (271). 128 addresses are tagged with database ransom activity (ATT&CK T1485 and T1486); all of them hit the MongoDB decoy.
In 24 hours the fleet recorded 103,583 credential pairs. The most-tried passwords are the familiar defaults (123456, password, 123456789) followed by complexity-rule variants such as Pa$$w0rd, Password@123 and Qwerty123 – passwords written to pass a policy rather than resist a guess.
The 454 distinctive request paths in paths-7d.txt lean towards secrets: 82 contain .env and 55 target /.git. The 16 user agents in user-agents-7d.txt are more telling. Eight are VPN client strings – Cisco AnyConnect, Citrix Receiver and Workspace, F5 Access, FortiSSLVPN, Palo Alto GlobalProtect and Pulse Secure – sent by tools posing as remote-access clients while they guess VPN logins. A further 79 listed addresses are tagged with VPN-login brute force. Addresses whose only activity is VPN-login guessing are mostly residential and mobile, so the feed keeps them out of its top block lists.
§ 07. Method and limits
Every number above comes from public files at feeds.offseq.com for the snapshot of 13 November 2026 (build bf44e78b404586cd): stats.json for the 24-hour observed totals, attackers-7d.csv for the listed set (VPN and Tor exits removed), cves-7d.csv, history.json, passwords-top100.txt, paths-7d.txt and user-agents-7d.txt. KEV dates come from CISA's catalog version 2026.11.13. EPSS percentiles come from the FIRST API on 14 November 2026. PHP2Shell timings come from OffSeq's earlier field report on its first 48 hours.
The fleet uses decoys, so it records attempts, not successful compromises. Country and network come from registration data. The listing rules drop activity seen by fewer than 3 sensors, so attacks aimed at a single target are under-represented.
The data is free to use under CC BY 4.0 at feeds.offseq.com.
§ 08. Observation timeline
- 2026-10-2116:21 UTC – first PHP2Shell session upload-progress attempt on the fleet
- 2026-10-22First day in the feed's published history
- 2026-10-26CISA adds CVE-2026-109618 and CVE-2026-109619 (PHP2Shell) to KEV
- 2026-11-03Listed attackers over 7 days peak at 3,974
- 2026-11-13Snapshot day: 1,693 listed attackers, 10 exploited CVEs
References
- Mirage Feed – data files, listing rules and history (CC BY 4.0) (opens in a new tab) (https://feeds.offseq.com)
- CISA – Known Exploited Vulnerabilities catalog (version 2026.11.13) (opens in a new tab) (https://www.cisa.gov/known-exploited-vulnerabilities-catalog)
- FIRST – EPSS API (opens in a new tab) (https://api.first.org/epss/)
- NVD – CVE-2025-5777 (Citrix Bleed 2) (opens in a new tab) (https://nvd.nist.gov/vuln/detail/CVE-2025-5777)
- NVD – CVE-2017-9841 (PHPUnit eval-stdin) (opens in a new tab) (https://nvd.nist.gov/vuln/detail/CVE-2017-9841)
Credit
Compiled by OffSeq from the public Mirage Feed snapshot of 13 November 2026, enriched with the CISA KEV catalog and FIRST EPSS scores of 14 November 2026.
More research
PHP2Shell in the wild: the first 48 hours on a honeypot fleet
PHP2Shell (CVE-2026-109618 and CVE-2026-109619) is unauthenticated remote code execution in the PHP runtime, leaked before its patch on 21 October 2026. OffSeq's Mirage honeypot fleet captured the first forty-eight hours of exploitation: nine source addresses, two attack shapes, and a tool that names itself in its own HTTP boundary.
Mirage field reportOct 22, 202610 min read
J@IL-GPT: an infostealer in a fake ChatGPT tool
A fake ChatGPT 4.5 tool distributed through DarkForums and GitHub contains a Windows infostealer. Its MSI and EXE packages conceal identical collection code, a browser-injection DLL, encrypted uploads and a conditional persistence branch.
Field investigationSep 6, 202615 min read
UpdraftPlus: PHP backdoor delivery attempts via CVE-2026-10795
A forged UpdraftPlus management request delivered a PHP backdoor to a Mirage decoy on 6 September. The retained campaign contains 20 plugin-upload attempts and 76 requests checking exact payload paths. Static analysis identifies command execution, file management and upload functions in the complete recovered sample.
Mirage field reportSep 6, 202613 min read