A Backdoor With No Server of Its Own: UAT-11587, Antino and Microsoft 365 as the Control Channel
Cisco Talos has documented eleven months of China-nexus espionage against government and policy bodies in eight Asian countries, including the Philippines, Cambodia, Thailand and Myanmar. The implant takes its orders from an Outlook mailbox and moves files through OneDrive.

What Talos documented
On 30 September, Cisco Talos published research by Ashley Shen on a China-nexus actor it tracks as UAT-11587 and a backdoor it names Antino [1]. Talos identified “at least 10 confirmed and five probable affected institutional environments, plus one additional intended target”, and “approximately 350 compromised endpoints across eight countries” [1]. The countries are Taiwan, India, the Philippines, Cambodia, Pakistan, Thailand, Myanmar and Syria [1].
Activity runs from September 2025 to July 2026. Early waves used Philippines-themed lures, campaigns widened in January 2026, and operations “accelerated between March and early June” [1]. Talos records a surge “on June 8 and 9, when Talos identified around 57 newly observed endpoints associated with India” [1].
One note on figures. Some secondary coverage described sixteen affected entities. Talos’s own count is the one above, and endpoint totals and organisation totals measure different things [1][2].
Why this target list matters here
The victim set is government and policy bodies, universities, think tanks and civil society organisations working on foreign affairs and international security [1][2]. Four of the eight countries are in Southeast Asia, and the lure themes Talos lists include Taiwanese government policy, Indo-Pacific affairs and maritime issues [3]. These are the institutions that host and attend regional security events.
An authentication result is not a verdict
The delivery tradecraft is the part most organisations can act on immediately. Talos describes the actor exploiting “the distinction between the SMTP envelope sender and the visible From header”, using a commercial mail provider with an attacker-controlled domain as the envelope sender “while the RFC5322 From header displayed the identity of the organization being impersonated” [1].
The consequence: “SPF passed” for the attacker’s domain, but “this result authenticated only the envelope-sender domain, not the sender displayed to the recipient. DMARC detected that the envelope and visible sender domains were not aligned and returned a failure” [1].
So the controls worked and reported correctly. What decided the outcome was enforcement. RFC 7489 defines this alignment check and what each policy value asks of a receiver: under “none” the domain owner “requests no specific action be taken regarding delivery of messages”, while under “reject” the owner “wishes for Mail Receivers to reject email that fails the DMARC mechanism check”, with rejection that “SHOULD occur during the SMTP transaction” [8].
A receiving organisation that publishes a policy but does not act on inbound alignment failures gains a log entry and nothing else. That is the gap this campaign walked through.
A picture of an attachment
There was nothing to scan. Talos found the actor “replicated the styling of Gmail’s attachment card using four inline PNG images embedded as Base64-encoded MIME parts”, and that “the entire attachment card was wrapped in an anchor tag pointing to an attacker-controlled URL” [1]. The recipient saw what looked like a document in their own client’s visual language and clicked a link.
Attachment sandboxing has no role here, because no attachment exists. The detection surface is the link and the user’s expectation: what looks like an attachment inside the message body is an image.
Living off signed binaries
Talos lays out five stages: an HTA executed by mshta.exe, a JScript downloader delivered from cloud storage, a .NET deserialization chain driven from that script, a loader assembly that fetches a decoy document and a sideloading bundle, and the implant [1].
The last stage is where detection belongs. Talos states that GatherOsState.exe “is a legitimate Microsoft-signed Windows Assessment and Deployment Kit (ADK) binary that was abused for DLL sideloading” and that “the attacker placed the Antino backdoor file slc.dll alongside the signed executable” [1]. Microsoft documents the ADK as a kit to “customize Windows images for large-scale deployment” [10] — a deployment-engineering toolkit, not something belonging in a user-writable directory on a policy officer’s laptop.
The technique is older than the campaign: a public hijack-library index added the pairing on 3 October, the contributor noting it is “a long-standing and reproducible technique” that “predates this campaign by years” [9]. So the useful detection is generic — a signed ADK binary executing from an unexpected path with a local slc.dll present.
Antino also avoids spawning PowerShell directly. Talos reports that it “abuses the Windows Scripted Diagnostics framework to execute attacker-controlled PowerShell through legitimate Windows components” [1]. That puts the parent process where most PowerShell alerting does not look.
The control channel is the victim’s own tenant
Antino is written in Rust and carries no control address of its own; it talks to Microsoft Graph. Talos describes the implant pulling “commands from the threat actor’s Outlook mailbox folder every 10 seconds”, request and response messages distinguished by subject prefixes, while OneDrive paths carry registration, heartbeats, staged tooling and exfiltrated files [1][3].
Microsoft documents Graph as “a single endpoint, https://graph.microsoft.com, to provide access to rich, people-centric data and insights in the Microsoft cloud”, spanning Outlook and OneDrive among others [6]. That is the whole problem: the destination is the hostname the organisation’s own productivity software talks to all day, so egress filtering, domain reputation and newly-registered-domain logic return nothing.
MITRE classifies this as bidirectional communication over a web service, where “adversaries may use an existing, legitimate external Web service as a means for sending commands to and receiving output from a compromised system”; its detection guidance points at behaviour rather than destination [5].
In practice the signal is the pairing. A Graph call is unremarkable; a Graph call from a process that is not a mail or sync client, on a fixed ten-second cadence, from a host that also ran a signed deployment binary out of place, is not. SOC Prime’s detection set is built on such pairings rather than on addresses [4].
What Talos did not claim
Two confidence statements are worth preserving. Talos assesses “with high confidence that UAT-11587 is China-nexus”, and says that judgement rests on “the totality of corroborating technical and operational evidence, rather than any single indicator” [1]. Supporting evidence includes Simplified Chinese document metadata, a UTC+08:00 time zone in phishing headers, and ten build outputs referencing a China-focused Rust package mirror — about which Talos notes that “the service’s public accessibility does not reveal the developer’s location” [1].
On the relationship to Jewelbug, which Symantec’s Threat Hunter Team described in August as running espionage and cryptocurrency fraud side by side [7], Talos is restrained: it observed overlaps but “could not independently verify a connection”, tracks UAT-11587 “as a separate activity set”, and rates the relationship at low confidence [1]. A brief that merges the two names loses that distinction and overstates what is known.
What to hunt
- Enforce inbound DMARC alignment failures rather than logging them, and check what your gateway does with a message whose envelope and visible sender domains differ [1][8].
- Hunt for mshta.exe or Windows Script Host retrieving remote content from cloud storage or content-delivery hosts your estate has no reason to use [1][4].
- Alert on any ADK binary executing outside its install path, and on an slc.dll in a user-writable directory [1][9][10].
- Look for PowerShell whose parent is the scripted-diagnostics host rather than a shell or management agent [1][4].
- Baseline which processes are expected to call Microsoft Graph and alert on the others, with regular-interval polling as the strongest signal [1][5][6].
- Review Microsoft 365 consent records for unexpected application authorisations against mail and file scopes [4].
The durable point is not the implant. An espionage operation against policy institutions in this region ran for eleven months using a signed Microsoft binary for execution, an image of an attachment for delivery, and the victim’s own cloud productivity endpoint for control. Each choice removes a detection opportunity most programmes still depend on, and each is answered by behavioural pairing rather than a list of addresses.
Sources
Every R3KONX article cites its primary material. 10 sources, in order of first citation. Links open the original publication.
- China-nexus UAT-11587 targets government and policy organizations across Asia with Antino backdoor Cisco Talos · 2026-09-30
- Antino Backdoor Uses Outlook and OneDrive for C2 in China-Nexus Espionage Campaign The Hacker News · 2026-10-01
- China-Linked Hackers Use Antino Backdoor in Asia eSecurity Planet · 2026-10-02
- UAT-11587 Targets Asian Government and Policy Organizations with Antino Backdoor SOC Prime · 2026-10-01
- T1102.002: Web Service: Bidirectional Communication MITRE ATT&CK · 2026-10-04
- Microsoft Graph overview Microsoft Learn · 2026-10-04
- Jewelbug: APT Group Runs Espionage and Crypto Fraud Operations Side by Side Symantec Threat Hunter Team, Broadcom · 2026-08-13
- RFC 7489: Domain-based Message Authentication, Reporting, and Conformance (DMARC) IETF · 2015-03-01
- Add gatherosstate.exe as a vulnerable executable for slc.dll (pull request 197) HijackLibs · 2026-10-03
- Download and install the Windows ADK Microsoft Learn · 2026-10-04
Researched and written by the R3KONX analysis desk from the cited primary material. Counts, dates and technical description are Talos's as published on 30 September; where secondary reporting states a different victim count, Talos's own figures are used and the discrepancy is noted. No indicators of compromise are reproduced here, and the attribution and actor-overlap confidence levels are quoted as Talos worded them rather than restated as fact. Corrections to event@r3konx.asia.
