Skip to content
NewCybercrime Report 2026 · Eight Chapters · A Post Is Not An Incident → Explore The ChaptersCybercrime Report 2026 → Explore
All full reports
Full reportCallback infrastructureMonograph 17 / 1917 min read

Out-of-Band Callback Fabrics: Interactsh, Burp Collaborator, DNSLog and a Cloud Cluster

How self-hosted callback servers turn blind server bugs into signals, and when they sit next to attacker tooling

By
Odin by Brandefense
Published
19 November 2026
Reading time
17 min read
Edition
Version 1.1 · TLP:CLEAR
Scope
Prepared November 2026

1. Executive Summary

In short

In a two-week observation window ending 18 November 2026, Odin by Brandefense mapped 1,794 out-of-band callback endpoints on 1,052 hosts. Of these, 757 are confirmed or suspected Interactsh, Burp Collaborator, DNSLog and workbench nodes. Most are dual-use testing servers. But 18 hosts share an IP with other offensive tools, and four offensive services appeared in one Alibaba Cloud subnet on the same day. Block direct DNS, HTTP and SMTP egress from internal servers and route their lookups through Protective DNS.

Many serious server bugs give no visible answer. A blind server-side request forgery (SSRF) or a blind SQL injection runs quietly in the back end. To confirm such a bug, testers and attackers both use Out-of-Band Application Security Testing (OAST) servers. The probe makes the target server call out to a domain the tester controls, by DNS, HTTPS or mail. A callback shows the server can be reached and can reach the internet.

This report maps self-hosted OAST servers, shows where they share hosts with command-and-control (C2) tools, and walks through one tight cluster of offensive services in Alibaba Cloud.

1

Most callback servers are dual-use, but 18 share hosts with offensive tools.

We count 638 Interactsh, 109 private Burp Collaborator and 8 DNSLog nodes, plus 2 co-hosted offensive workbenches. Most look like ordinary testing servers. Across the fleet, 40 hosts expose four or more listener ports or extra services. Of these, 18 put a callback listener on the same IP as another offensive framework: 8 corroborated and 10 candidates.

2

Four unlinked offensive services sat in one Alibaba Cloud subnet.

On 18 November 2026, Odin first observed a DNSLog collector, a VShell WebSocket listener, a CyberStrikeAI engine and a HexStrikeAI tool bridge in 8.148.0.0/16. All four sit on separate IPs. Nothing links the four servers, so treat them as separate hosts in a busy cloud range.

3

Exposed attacker files tie OAST callbacks to exfiltration scripts and Java exploit templates.

Open web directories held 40 callback records. Of these, 18 sit on three confirmed offensive servers, in Python DNS exfiltration scripts and Java exploit templates. Another 15 records were incoming probes logged on a scanned third-party web application, not attacker files.

Three Decisions for Leaders

  1. Restrict outbound DNS, HTTP and SMTP from internal application and database servers. Back-end web servers, middleware containers and databases should resolve names only through internal Protective DNS forwarders.
  2. Watch recursive DNS for callback-style lookups. Alert when a server process queries a public callback zone or resolves a 26 to 36 character random-looking label.
  3. Block only callback servers with independent offensive evidence. Raise priority when a listener shares a host with a C2 framework or an offensive workbench. Send standalone listeners to analyst review, because authorized testers run them too.

The Study in Numbers

NumberWhat it is
1,794OAST endpoints on 1,052 IPv4 hosts, after 356 AWS load-balancer reflection collisions were set aside
757Confirmed or suspected callback nodes (755 listeners plus 2 co-hosted offensive workbenches)
638Interactsh endpoints among them (523 confirmed, 112 suspected, 3 on dynamic ISPs)
109 and 8Private Burp Collaborator endpoints (109) and DNSLog listeners (8)
4Complementary offensive services first observed in 8.148.0.0/16 on 18 November 2026
40 and 18Multi-service hosts (40) and hosts where a listener shares an IP with another offensive framework (18)
40Callback records found in exposed web directories, 18 of them on confirmed offensive workbenches
9 of 28Selected indicators ready to block; the other 19 are watchlist, review or reference entries

How to use this report.

Security leaders can read Chapter 1, Chapter 6 and Chapter 7. Network, cloud and database teams should read Chapters 2 to 5 and use the checklist in Chapter 7. Detection engineers and hunters will find the rule logic in Chapter 8 and the indicators in Chapter 9.

2. The Callback Fleet and the Four-Role Model

In short

The 1,794 endpoints break down into 757 confirmed or suspected nodes and 1,039 candidates. Most are standalone testing servers, so the useful signal is co-hosting with C2 and reconnaissance tools.

OAST servers fix a visibility problem. Many server-side flaws return no output in the HTTP response. An OAST server runs its own authoritative DNS name server and a wildcard TLS listener. When the target server looks up or fetches a tagged address, the OAST server logs it. That turns a silent bug into a record the tester can read.

For defenders, this creates a choke point. The callback has to leave the internal server. Egress filtering and DNS logging can see it. External listener data alone does not show whether any target was breached.

An analytical model of four roles

Figure 1 is an analytical model, not an observed operation. It shows how four cloud roles could link an attack from probe to implant.

Figure 1: Four automated cloud roles link blind vulnerability verification to implant C2
Figure 1: Four automated cloud roles link blind vulnerability verification to implant C2
RoleExample toolsWhat it does in the model
AI offense orchestratorCyberStrikeAI, HexStrikeAI, MCP command-line bridgesQueues scans, reads callback results and picks the next task
Blind probe dispatcherNuclei YAML templates, SSRF, XXE, JNDI and SQLi payloadsTags outbound probes with a unique correlation ID
OAST callback collectorInteractsh, Burp Collaborator, DNSLog on 8000Captures DNS, TLS and SMTP callbacks that confirm blind reach
Implant callback handlerVShell WebSocket, Sliver on 443, ARTEX on 8787Would hold implant sessions from servers that answered, over a TLS WebSocket channel

MCP, the Model Context Protocol, lets a language model call external tools. Nothing in our data shows one operator running all four roles. The model is a way to read the co-hosting and the subnet cluster in later chapters.

Fleet counts by platform

Each endpoint is an IP and port pair.

PlatformEndpoints (IPs)ConfirmedSuspected + dynamic ISPConfirmed or suspectedCandidates
Interactsh1,245 (700)523 (6 with open directory)112 + 3638607 (6 with open directory)
Burp Collaborator516 (319)6245 + 2109407
DNSLog33 (33)17 + 0825
Offensive open-directory workbenches2 (2 shared)2020
Total1,796 (1,052)588164 + 57571,039

The total row counts 1,796 records: the 1,794 OAST endpoints plus the 2 workbenches. A further 356 AWS Application Load Balancer endpoints echo any header they receive. They look like OAST servers but are not, so they are left out of every count.

Where the fleet is hosted

Hosting splits by platform. Interactsh and Burp Collaborator sit mostly on Western cloud providers. DNSLog sits mostly in China.

PlatformTop providers (endpoints)Top countries (endpoints)
InteractshAmazon.com, Inc. (360), DigitalOcean, LLC (180)United States (575), Germany (145)
Burp CollaboratorAmazon.com, Inc. (142), DigitalOcean, LLC (108)United States (180), Germany (84)
DNSLogTencent (11), Hangzhou Alibaba Advertising Co.,Ltd. (9)China (20), Hong Kong (6)

Co-hosting: the strongest signal

Across the 1,052 hosts, 40 IPv4 addresses expose several services. Of these, 22 are standalone listener matrices with four or more callback ports. The other 18 put a callback listener on the same IP as a C2, reconnaissance or open-directory service. Eight of those 18 are corroborated and listed below. One of the eight has its address withheld, because its Sliver port is only suspected. The 10 candidate co-locations are withheld.

Host (country)Hosting networkCallback listener portsCo-hosted offensive service
Address withheld (US)NetLab GlobalInteractsh 465, 587, 18080Suspected Sliver on 443, plus an offensive open directory on 8000 (1,417 files)
49.232.168[.]207 (CN)Shenzhen TencentDNSLog 8000 (confirmed)VShell WebSocket C2 listener on 5030
101.33.231[.]97 (CN)Shenzhen TencentInteractsh 443Cobalt Strike listener on 8080
43.153.118[.]75 (US)Shenzhen TencentInteractsh 8300VShell WebSocket listeners on 8084 and 8094
23.94.214[.]197 (US)HostPapaInteractsh 465, 587, 443ARTEX post-exploitation listener on 8787
150.158.9[.]124 (CN)Shenzhen TencentDNSLog 8000ARL (Asset Reconnaissance Lighthouse) on 5003
51.15.224[.]47 (FR)Scaleway SASInteractsh 80, 443, 465, 587, 3000ezXSS blind XSS payload manager on 9443
49.51.187[.]59 (US)Shenzhen TencentInteractsh 80 (confirmed)Offensive open-directory workbench on 8090 (67 files)

3. How the Three Platforms Look on the Wire

In short

Interactsh, Burp Collaborator and DNSLog each leave their own headers, routes and TLS fingerprints. Multi-port listeners on 53, 80, 443, 465, 587 and 8000 catch a callback over whichever protocol the firewall leaves open.

Public shared callback zones such as oast[.]pro and oastify[.]com are well known. Many Protective DNS services already sinkhole them. To avoid those blocks, operators can run private collectors on their own domains and cheap cloud servers. Authorized red teams do the same, for the same reason.

Figure 2: Authoritative DNS and wildcard TLS listeners convert blind bugs into callback logs
Figure 2: Authoritative DNS and wildcard TLS listeners convert blind bugs into callback logs

Self-hosted Interactsh

ProjectDiscovery's interactsh-server runs DNS, HTTP, HTTPS and SMTP collectors under one wildcard domain. It answers with an X-Interactsh-Version header or Server: Interactsh-Server.

FeatureValue
Endpoints (IPs)1,245 (700)
Most common ports80 (374), 587 (303), 443 (267), 465 (216)
TLS fingerprint229 of 267 HTTPS nodes on 443 (85.8%) share the Go TLS JARM 29d29d00029d29d00041d...
Client polling route/poll?id= with a secret parameter

A JARM value is a fingerprint of how a server answers TLS handshakes. A shared JARM points to the same software stack, not the same owner. Standalone Interactsh matrices with four ports and no other tooling stay on the watchlist.

Private Burp Collaborator

PortSwigger lets customers run a private Collaborator server. It is set up with a JSON file (serverDomain, eventCapture, polling) and answers with Server: BurpCollaboratorServer.

FeatureValue
Endpoints (IPs)516 (319)
Most common ports443 (213), 80 (199), 465 (34), 587 (27), and 8443, 3380, 33443 (6 each)
TLS fingerprint236 of 259 TLS nodes (91.1%) share the Java JARM 28d28d28d00028d00042d...

Several standalone Collaborator servers bind custom certificate domains, and one hub listens on 18 ports. None shows co-hosted C2 tooling, so they stay on the watchlist and their addresses are withheld.

DNSLog

DNSLog pairs an authoritative DNS listener on 53 with an HTTP API. The API runs on 8000 on 18 of 33 endpoints (54.5%).

Two route families exist. dnslog[.]cn-compatible panels expose /getdomain and /getrecords (or the .php forms). BugScanTeam's DNSLog uses /api/ routes instead. Suricata SID 9517006 covers only the first family.

Two DNSLog hosts share an IP with offensive tools: 49.232.168[.]207 with VShell on 5030, and 150.158.9[.]124 with ARL on 5003.

4. From Blind Probe to Callback: What Exposed Directories Held

In short

Exposed web directories held 40 callback records. Eighteen sit on three confirmed offensive servers, one is a standalone client log, six are template mirrors, and fifteen were logged on a scanned third-party target.

In manual testing, an analyst pastes one Collaborator address into a request and checks for callbacks by hand. In automated work, scanners such as Nuclei, HexStrikeAI and CyberStrikeAI poll the OAST API themselves. Every probe carries a unique correlation ID, so each callback maps back to one probe.

Figure 3: Out-of-band DNS and HTTPS callbacks bridge blind probes to WebSocket C2
Figure 3: Out-of-band DNS and HTTPS callbacks bridge blind probes to WebSocket C2

Figure 3 is a reference sequence built from public protocol behaviour. Steps 5 and 6 show a possible follow-on stage. No victim-side logs are part of this study.

Four blind bug classes that call out

Firewalls often block inbound traffic but allow recursive DNS on UDP 53 or outbound TLS on 443. That lets a blind bug call out by one of four routes.

Bug classHow the callback happens
Blind SSRFA web app, webhook previewer or PDF generator fetches an attacker-supplied https://<correlation-id>.<oast-zone> URL. This reveals the server's egress IP and HTTP client.
Blind SQL injectionWith errors hidden, a query calls database network functions such as xp_dirtree (SQL Server) or UTL_HTTP and UTL_INADDR (Oracle). These trigger a DNS lookup.
Blind XXE and JNDIAn XML parser fetches an external DTD, or a Java logging sink makes an anonymous LDAPv3 bind on 389 or 1389.
Asynchronous command executionA worker runs nslookup, curl or wget on a multi-label name such as <hex-chunk>.<correlation-id>.oast.pro. This confirms execution or leaks short strings.

The 40 callback records

The 40 records use 15 distinct callback names or wildcard patterns across 8 public shared OAST zones. The 18 records on confirmed workbenches use 8 distinct callback names.

SourceFilesRecords (distinct)What they show
144.172.99[.]199:8080, confirmed workbench/dns_exfil.py, /dns_mass_exfil.py, /assemble_dnslog.py, /dbg*.log12 (5)DNS exfiltration scripts using oast[.]pro, oast[.]site, oast[.]me and v9gcgt.dnslog[.]cn
47.101.51[.]235:8888, confirmed workbench/EvilTemplate.java, /D2.java, /ET.java, /M.java, /M2.java5 (2)Java exploit templates calling 57v1we22.dnslog[.]cn and 25cot090vm.dnslog[.]cn
45.145.228[.]16:8080, confirmed workbench/memshell/ExploitScriptEngineFactory.java1 (1)Java memory-shell template calling wiiesroglrl1t9t7ljwhkjvo1f7cv1.oastify[.]com
Standalone host (withheld)/att_interactsh.out1 (1)Captured Interactsh client output for an oast[.]pro name; watchlist
Template mirrors on 5 hosts (withheld)Nuclei templates and vulnerability wiki pages6 (5)Public scanner templates using *.oast[.]online, *.oast[.]pro, *.interact[.]sh and *.burpcollaborator[.]net
Scanned third-party web application (withheld)A Laravel session file15 (1)Incoming scanner probes using an oast[.]site name, logged by the target

These are staged files. They show intent and tooling. They do not show that any probe or script worked against a real target.

The last row matters for hygiene. That host is a probable target of scanning, not an attacker. It is withheld and should not be blocked.

5. Four Services in Alibaba Cloud 8.148.0.0/16

In short

On 18 November 2026, Odin first observed four complementary offensive services in Alibaba Cloud 8.148.0.0/16. They sit inside a wider set of 17 endpoints on 13 hosts in the same range.

Single-host indicators miss how cloud infrastructure clusters. All four services sat in one Alibaba Cloud range (Hangzhou Alibaba Advertising Co.,Ltd., CN).

Figure 4: Four complementary offensive services were first observed in 8.148.0.0/16 on the same day
Figure 4: Four complementary offensive services were first observed in 8.148.0.0/16 on the same day

The four services in order

PortServiceFramework labelFingerprint and role
8000DNSLogSuspectedHTTP /getdomain and /getrecords callback collector
443VShellSuspectedGo TLS JARM 29d29d16d..., WebSocket C2 listener; certificate domain withheld
8080CyberStrikeAIConfirmedCN=CyberStrikeAI, JARM 40d40d40d..., autonomous offense engine
5000HexStrikeAISuspectedREST MCP bridge that hands command-line scanners to a model

The four IPs fall between 8.148.195.0/24 and 8.148.233.0/24. Each exposes a different port and a different role: callback collection, WebSocket C2, AI orchestration and MCP tool execution. Their addresses are withheld.

Nothing does. We found no shared certificate, cross-host control link or provisioning file on any of the four servers. Alibaba Cloud ranges host many unrelated customers, including testers. The roles fit together on paper, but that is not evidence of one operator.

Because nothing proves shared control, all four hosts stay on the watchlist. They are not on the block list.

The rest of the /16

Beyond the four, the range holds 13 more records on 9 sibling IPs. That makes 17 endpoints on 13 IPs, after setting aside 2 unrelated 3306 false positives.

Only one host in the whole range has the same-host evidence we need for blocking. 8.148.31[.]135 runs a confirmed Metasploit Meterpreter listener on 4446 and 443, plus a Sliver listener on 8888. Its certificate domains are withheld.

The 8 other siblings carry 10 endpoints, all withheld:

ServicePortNote
CyberStrikeAI8080CN=CyberStrikeAI; carries an earlier first-seen date of 2 November 2026
PentAGI8443Single listener
DNSLog (TLS)8443Standalone
ARTEX8787Candidate
Hashtopolis (password-cracking manager)8443Candidate
Sliver8443Candidate
VShell8090Candidate
mitmproxy18081, 18082, 18084One host, three ports

6. Telling Testers from Attackers

In short

Interactsh and Burp Collaborator are everyday tools for contracted testers and bug-bounty researchers. Blocking every listener breaks legitimate work, so only 9 of 28 selected indicators are block-ready.

Figure 5 shows how far the confirmed and suspected set shrinks from the full endpoint count for each platform.

Figure 5: 757 confirmed or suspected OAST nodes stand out from 1,794 endpoints
Figure 5: 757 confirmed or suspected OAST nodes stand out from 1,794 endpoints

What earns a block

A callback server goes on the block list only with independent offensive evidence. Two kinds qualify. The first is a C2 or reconnaissance framework on the same IP. The second is an exposed directory holding offensive scripts or exploit templates.

On that basis, 9 selected indicators are block-ready. That is 7 co-hosted callback or C2 nodes and 2 offensive workbenches. The 7 are six callback hosts from Chapter 2 plus 8.148.31[.]135 from Chapter 5. One more co-hosted host is withheld, because its Sliver port is only suspected and a red team cannot be ruled out.

What stays off the block list

GroupSizeWhy not blockAction
AWS load-balancer reflections356 recordsThey echo headers and only look like OAST serversExclude before any analysis
Standalone listeners and single-service /16 hosts7 standalone OAST and 7 /16 entries in the selected setNo independent offensive evidence; often authorized testersWatchlist; enrich with passive DNS
Consultancy Collaborator server1Carries an identifiable testing-firm certificate and no C2Analyst review; check test scope
Listeners on dynamic subscriber lines5 endpoints (3 Interactsh in Vietnam and China, 2 Burp Collaborator in the UK)Home and small-business addresses change handsShort watchlist that expires after 24 hours
Scanned third-party web application1 host, 15 recordsIt logged incoming probes; it is a likely targetWithhold and exclude from threat lists
Your own approved scannersNot countedInternal DAST and red-team runners use the same toolsAllowlist by exact host, for scheduled windows

The same caution applies to the subnet cluster. Shared cloud ranges also host independent security researchers and commercial testing tenants.

7. What Defenders Should Do

Within 24 hours

  • Block the 9 block-ready endpoints in Chapter 9 at the perimeter, proxy and server egress.
  • Hunt past DNS and HTTP logs for contact with those hosts and the eight workbench callback names in Chapter 4.
  • Put the 18 watchlist and review entries in enrichment tables, not in firewall blocks.

Within 72 hours

  • Block direct outbound UDP and TCP 53, HTTP and HTTPS (80, 443, 8000), SMTP (465, 587) and LDAP (389, 1389) from web, middleware and database subnets, except to explicit allowlists.
  • Force all server DNS through internal Protective DNS forwarders that sinkhole oast.*, interact[.]sh, oastify[.]com, burpcollaborator[.]net and dnslog[.]cn.
  • Register every internal DAST or red-team scanner IP and private Collaborator or Interactsh domain in a versioned allowlist before each test window.

Within 7 to 30 days

  • Disable or restrict database network functions: xp_dirtree and xp_subdirs on SQL Server, UTL_HTTP and UTL_INADDR on Oracle, and COPY ... FROM PROGRAM on PostgreSQL.
  • Deploy Suricata SIDs 9517001 to 9517010, the Sentinel server-egress query and the three Sigma rules.
  • Alert on server lookups of 26 to 36 character random-looking labels.
  • Recheck TLS certificate subjects and JARM values on elastic cloud IPs at regular intervals, since addresses get reassigned.

8. Detection Rules

The full rules are in the Detection rules tab. They use explicit host allowlists rather than wildcard exclusions.

Rule fileWhat it catches
odin_oast_interactsh_burpcollab_dnslog.rulesSuricata 7: ten signatures for callback DNS lookups, OAST server headers, DNSLog and Interactsh polling, and anonymous LDAP callbacks
sentinel_internal_server_oast_egress_hunt.kqlMicrosoft Sentinel and Defender XDR: application and database server processes that reach OAST zones or the corroborated hosts, with follow-on activity
sigma_server_tier_oast_dns_and_http_callback.ymlSigma (3 rules): server OAST lookups, database or web workers starting lookup tools, and unapproved OAST or offense bridge daemons
odin_oast_payload_and_provisioning_scripts.yarYARA (4 rules): Interactsh and Collaborator server configs, DNSLog polling files, and multi-stack cloud provisioning files

How the Suricata signatures work

SIDLooks forWhere to deploy
9517001DNS query to .oast[.]pro, .live, .site, .online, .fun or .meDNS forwarder egress
9517002DNS query for a long label under .oastify[.]comDNS forwarder egress
9517003HTTP response header X-Interactsh-Version or Server: Interactsh-ServerTLS inspection point
9517004HTTP response header BurpCollaboratorServerTLS inspection point
9517005Outbound anonymous LDAPv3 bind to 389, 1389 or 10389Perimeter TCP egress
9517006HTTP request to /getrecords or /getrecords.php (BugScanTeam /api/ routes not covered)TLS inspection point
9517007Interactsh client polling with /poll?id= and &secret=TLS inspection point
9517008DNS query for a 20 to 36 character label under .interact[.]shDNS forwarder egress
9517009DNS query for a long label under .burpcollaborator[.]netDNS forwarder egress
9517010DNS query under .dnslog[.]cnDNS forwarder egress

Bypass SIDs 9517001 to 9517004 and 9517007 to 9517010 only for documented scanner source IPs during scheduled test windows. Treat any SID 9517005 alert as high priority. Treat the DNS and header SIDs as high priority when they fire from a server subnet.

How the Sentinel query works

The query looks back 14 days. It watches network events from server processes such as java, w3wp.exe, sqlservr.exe, postgres, mysqld, node and dotnet. It skips listed scanner devices.

SignalFires when
Callback followed by follow-onA server process reaches an OAST zone or a correlation-style name, then within 10 minutes starts a shell or curl/wget, or connects to a corroborated C2 or workbench endpoint
Standalone server callbackA server process reaches an OAST zone or correlation-style name
Egress to a corroborated hostA server process connects directly to one of the 11 corroborated hosts in the query's list
Port-only reviewA server process reaches a public IP on 465, 587, 389, 1389 or 8000 with no OAST indicator and no approved relay

Fill in the approved scanner hostnames and the approved SMTP, LDAP or webhook relay IPs before use. The port-only signal is the weakest and is meant for review.

How the Sigma and YARA rules work

RuleFires when
Sigma 1: server OAST DNS queryA Windows web, middleware or database process queries an OAST zone, outside approved scanner hosts
Sigma 2: worker spawns lookup toolA database or web worker starts nslookup.exe, curl.exe, wget.exe or certutil.exe, outside health checks
Sigma 3: OAST or offense daemonA Linux host runs interactsh-server, --collaborator-server, dnslog, cyberstrikeai or hexstrikeai, outside sanctioned red-team hosts
ODIN_Interactsh_SelfHosted_Server_ConfigAn Interactsh server name appears with SMTP port settings or wildcard and listen settings
ODIN_BurpCollaborator_Custom_Server_ConfigCollaborator JSON keys such as serverDomain, polling and eventCapture appear together
ODIN_DNSLog_And_OAST_Polling_DescriptorA small poller file holds full http URLs for both /getrecords and /getdomain
ODIN_IaC_MultiStack_Offensive_Burst_ManifestA Terraform file declares an Alibaba Cloud instance and lists quoted roles such as cyberstrikeai, hexstrikeai, vshell or dnslog with ports

The YARA rules were checked against four sanitized, non-executable descriptors and against BIND9 and Terraform baselines, with no false positives. The provisioning rule is built from an illustrative example; no such Terraform file was recovered.

How to sort a host

What the host showsAction
A callback listener plus a C2 or recon framework on the same IPBlock
An exposed directory with offensive scripts or exploit templatesBlock
A standalone listener, or a single-service host inside a cloud clusterWatchlist and analyst review
A consultancy certificate with no C2Analyst review
A listener on a dynamic subscriber line24-hour watchlist
A third-party app logging incoming probesExclude from threat lists
A load-balancer reflectionSuppress
A documented internal scannerAllow

9. Indicators

The Indicators tab holds the 9 block-ready endpoints. The other 19 records in the selected set are withheld.

TypeCount
Block-ready endpoints (IP and port)9
Watchlist and review entries (addresses withheld)18
Load-balancer collision reference (withheld)1
Distinct callback names in confirmed workbench files8

Supporting fingerprints: the Interactsh JARM prefix 29d29d00029d29d00041d..., the Burp Collaborator JARM prefix 28d28d28d00028d00042d..., the VShell JARM prefix 29d29d16d..., the CyberStrikeAI JARM prefix 40d40d40d... and the certificate subject CN=CyberStrikeAI.

Block-ready endpoints

EndpointRoleCountryHostingEvidenceEarliest first seen on record
8.148.31[.]135:4446Metasploit MeterpreterCNAlibaba CloudMeterpreter on 4446 and 443, Sliver on 888813 Nov 2026
49.232.168[.]207:8000DNSLogCNTencentConfirmed DNSLog with VShell on 503013 Nov 2026
101.33.231[.]97:443InteractshCNTencentCobalt Strike on 808013 Nov 2026
43.153.118[.]75:8300InteractshUSTencentVShell on 8084 and 809413 Nov 2026
23.94.214[.]197:587InteractshUSHostPapaInteractsh on 443, 465, 587; ARTEX on 878713 Nov 2026
150.158.9[.]124:8000DNSLogCNTencentARL recon panel on 500313 Nov 2026
51.15.224[.]47:587InteractshFRScalewayFive-port Interactsh with ezXSS on 944313 Nov 2026
45.145.228[.]16:8080Offensive open directoryHKCloudie/memshell/ExploitScriptEngineFactory.java with an oastify[.]com callbackBefore this window
47.101.51[.]235:8888Offensive open directoryCNAlibaba Cloud/EvilTemplate.java and five dnslog[.]cn callbacksBefore this window

Records marked "Before this window" carry first-seen dates older than our observation. They are not the start of our observation, which covers the two weeks ending 18 November 2026.

The Sentinel query also lists two more corroborated hosts from earlier chapters: the workbench 144.172.99[.]199:8080 and the co-hosted workbench on 49.51.187[.]59:8090.

Withheld: the four /16 cluster hosts, the eight /16 siblings, and standalone Interactsh and Collaborator listeners with their certificate domains. Also withheld: the client log host, the five template mirrors, the consultancy Collaborator, the five subscriber addresses, the scanned Laravel target and the load-balancer collision host.

ATT&CK context

This activity maps to three MITRE ATT&CK tactics: Initial Access (blind probes and JNDI callbacks), Discovery (callback DNS lookups) and Command and Control (polling and co-hosted C2). See the Suricata table in Chapter 8 for which SID covers each.

10. Scope and Limits

  • Window. The findings cover a two-week observation window ending 18 November 2026. Some records carry earlier first-seen dates; we do not use those as a tracking start.
  • Dual-use by default. Interactsh, Burp Collaborator and DNSLog are standard testing tools. A listener alone says nothing about intent.
  • Candidates are leads. The 1,039 candidates are hunting leads. Do not block them.
  • Proximity is not control. Four services in one range show where they sat, not who runs them. No cross-host link was found.
  • Files are not outcomes. Workbench scripts and templates show tooling and intent. We did not see any of them used against a target system.
  • No victim view. This study looks at listeners from the outside. It has no logs from any target network.
  • Collisions removed. 356 AWS load-balancer reflections are excluded from every count.
  • Rule testing. Every published rule file loads without errors in its engine (Suricata 7.0.12, YARA-X, pySigma and Microsoft's Kusto.Language analyzer, checked on 20 November 2026). The rules were not replayed against live traffic, so test and tune them in your own environment before production use.

11. Sources

  1. MITRE ATT&CK Enterprise tactics. https://attack.mitre.org/tactics/enterprise/ (verified 19 November 2026)
  2. ProjectDiscovery. Interactsh repository. https://github.com/projectdiscovery/interactsh (verified 19 November 2026)
  3. PortSwigger. Deploying a private Burp Collaborator server. https://portswigger.net/burp/documentation/collaborator/deploying (verified 19 November 2026)
  4. BugScanTeam. DNSLog repository. https://github.com/BugScanTeam/DNSLog (verified 19 November 2026)
  5. NSA and CISA. Selecting a Protective DNS Service. https://www.nsa.gov/Press-Room/News-Highlights/Article/Article/2523771/nsa-and-cisa-release-cybersecurity-information-on-protective-dns/ (verified 19 November 2026)
  6. OWASP. Server-Side Request Forgery Prevention Cheat Sheet. https://cheatsheetseries.owasp.org/cheatsheets/Server_Side_Request_Forgery_Prevention_Cheat_Sheet.html (verified 19 November 2026)

What is…?

Short explanations of the tools and families named in this study.

All explanations

Frequently asked questions

What is a callback server, and why should I care?

Some of the worst server bugs are silent. A blind server-side request forgery or a blind SQL injection runs in the back end and shows nothing in the web page. To confirm such a bug, the tester plants a unique address in the probe. If the target server looks up or fetches that address, a callback server logs it.

Is every callback server a threat?

No. Contracted penetration testers and bug-bounty researchers run these tools every day. Blocking all of them would break legitimate work and add little. The signal that matters is company. Across the fleet, 40 hosts expose several services. On 18 of them, a callback listener shares an IP with another offensive framework. Eight of those are corroborated.

What sat in one cloud range?

On 18 November 2026, Odin first observed four offensive services in Alibaba Cloud's 8.148.0.0/16 range, on four separate servers. They were a DNSLog callback collector, a VShell WebSocket C2 listener, a CyberStrikeAI engine and a HexStrikeAI tool bridge. Each covered a different role.

Defender kit

Indicators of compromise

Verified adversary infrastructure and tool hashes from this study. Every value below is also discussed in the report.

IDTypeIndicatorPortContextAction
R17-IOC-01 IPv4 8.148.31[.]135 4446 Metasploit / Meterpreter handler Block
R17-IOC-02 IPv4 49.232.168[.]207 8000 dnslog Block
R17-IOC-03 IPv4 101.33.231[.]97 443 interactsh Block
R17-IOC-04 IPv4 43.153.118[.]75 8300 interactsh Block
R17-IOC-05 IPv4 23.94.214[.]197 587 interactsh Block
R17-IOC-06 IPv4 150.158.9[.]124 8000 dnslog Block
R17-IOC-07 IPv4 51.15.224[.]47 587 interactsh Block
R17-IOC-08 IPv4 45.145.228[.]16 8080 opendir_oast Block
R17-IOC-09 IPv4 47.101.51[.]235 8888 opendir_oast Block

19 records from this study are not published in this table: candidate, suspected and dual-use hosts, victim-related values, and hashes of exposed secrets or files.

Defender kit

Detection rules

The rule files exactly as printed in the report. Test them in your own engine before deployment.

Suricata odin_oast_interactsh_burpcollab_dnslog.rules 35 lines
Download
YARA odin_oast_payload_and_provisioning_scripts.yar 79 lines
Download
Sentinel KQL sentinel_internal_server_oast_egress_hunt.kql 89 lines
Download
Sigma sigma_server_tier_oast_dns_and_http_callback.yml 107 lines
Download