Nine labels, one cluster: the Silent Lynx reporting corrects itself
Cisco Talos and Recorded Future published joint reporting on 14 October on a government-focused espionage cluster across Central Asia and the CIS. The conference paper behind it also retracts an implant that earlier vendor reporting had named, which is the more useful finding.
What was published
On 14 October 2026 Cisco Talos and Recorded Future’s Insikt Group published coordinated reporting on a cyber-espionage cluster that has targeted government organisations across Central Asia and the Commonwealth of Independent States since at least 2024 [1][2]. Talos calls it UAT-6275 and assesses a Kazakhstan nexus at medium confidence [1]. Recorded Future calls it TAG-193 and assesses that the group likely operates from Kazakhstan [2]. The underlying work is a joint paper by Chi-en Shen of Talos and Julian-Ferdinand Vögele of Recorded Future, given at the Virus Bulletin conference in Seville on 14 to 16 October [3].
The targeting is consistent. Victims and lure themes span nine countries: Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, Uzbekistan, Armenia, Azerbaijan, Russia and Belarus [1][3]. The focus is government bodies handling diplomacy, economic development, agriculture and environmental management, with recurring interest in water and energy, including an Armenian environmental agency and Uzbekistan’s Smart Forestry initiative [1][2]. Talos ties activity to a Tajik entity after March 2026 treaty talks, and to a Belarusian entity before an April bilateral meeting [1].
Both teams flag the limit of that observation. Recorded Future states that the timing links do not demonstrate a connection or reveal what was sought, and that lure documents establish targeting interest rather than compromise [2]. The restraint matters, because the timing correlation is the detail most likely to be repeated without it.
Nine labels
The same activity, or activity sharing indicators with it, has been published under nine names, and the relationships each vendor claims are not identical.
- Seqrite Labs named Silent Lynx in January 2025 after two campaigns against the National Bank of the Kyrgyz Republic and the Kyrgyz Ministry of Finance, attributing it to a Kazakhstan-origin actor at medium confidence and assessing that it likely shares resources with YoroTrooper [4][5].
- Cisco Talos first reported YoroTrooper and in October 2023 assessed with high confidence that it likely consists of individuals from Kazakhstan, with infrastructure largely hosted in Azerbaijan to disguise that origin [9]. Talos now puts UAT-6275 and YoroTrooper in the same broader cluster but keeps them separate for want of direct overlap [1].
- Group-IB named ShadowSilk in August 2025 and assigned it a separate cluster despite toolset and infrastructure overlap with YoroTrooper, including shared PowerShell download URLs on a compromised Tajik government site [7].
- Kaspersky’s Tomiris, BI.ZONE’s Cavalry Werewolf, VMRay’s Hydra Saiga and Sturgeon Phisher have all been reported as overlapping [1][6], and the Virus Bulletin paper states that ShadowSilk and Hydra Saiga share nearly identical infrastructure [3].
For a defender the count is the problem. A detection rule keyed to one of these names does not inherit the indicators published under the other eight, and most of the relationships above are held open rather than resolved.
The correction
The most useful contribution of the Virus Bulletin paper is not a new actor profile. It is a retraction. The paper states that the SilentSweeper .NET implant described in earlier reporting reflects generic switches of a PowerShell-to-executable packer, not a distinct implant [3]. SilentSweeper was named in Seqrite’s November 2025 Operation Peek-a-Baku report, alongside Silent Loader and LAPLAS [6]. A named malware family is now a build artefact.
The same paper separates Chvost from Laplas as a distinct family on code and artefact differences, where earlier research grouped them, and introduces the earliest known Chvost sample, compiled June 2022 [3]. Talos lists Chvost, Laplas and PickleShell as the group’s reverse shells; Recorded Future lists LAPLAS, FoalShell variants and PowerShell reverse shells [1][2]. The paper adds PSReverso and SSDPShell [3].
The paper is also open about an internal inconsistency: it cites 76 GitHub commits in one section and 77 in the timing analysis used to infer operator working hours [3]. Talos gives 77 [1]. Numbers attached to an attribution argument are exactly where arithmetic should be auditable.
What the attribution actually says
Across every source the attribution stops short of a state. Talos assesses a Kazakhstan nexus at medium confidence, on infrastructure access, Russian-language artefacts, malware-testing submissions from Kazakhstan shortly after compilation, working hours consistent with UTC+5, and regional targeting [1]. The Virus Bulletin paper presents the same evidence as moderate confidence rather than definitive [3]. Recorded Future states plainly that the indicators do not establish organisational affiliation or state sponsorship [2].
Group-IB complicates this further, describing two subgroups in ShadowSilk, one Russian-speaking and handling malware development and initial access, one Chinese-speaking and handling network penetration and collection, with the depth of cooperation between them left uncertain [7]. Treating any of this as a single state programme goes beyond what has been published.
The counts disagree
Group-IB’s blog says over 35 victims were identified, primarily government in Central Asia, naming Kyrgyzstan, Uzbekistan, Tajikistan and Turkmenistan in the region and Pakistan and Myanmar in the Asia-Pacific [7]. Infosecurity Magazine’s account of the same investigation, supported by CERT-KG, says at least 35 [8]. Trade headlines carried 36. Three figures, one dataset. The Asia-Pacific entries matter here: a cluster framed everywhere as Central Asian has confirmed government victims in Pakistan and Myanmar, used Telegram bots to blend command and control with messenger traffic, and had stolen data appear on criminal forums [7][8].
What to act on
The delivery methods are ordinary, and that is the point. Macro-enabled Word documents, PowerShell downloaders wrapped as executables, shortcut files that run scripts hosted on GitHub, and password-protected archives sent from compromised government mailboxes [1][2][5]. Phishing from a genuine ministry address defeats sender reputation entirely.
- Patch archive utilities. The group’s first observed use of a vulnerability for delivery was WinRAR CVE-2025-8088, a path traversal used in April 2026 to place a reverse shell in the Startup folder [1][2][3]. The record carries CVSS 8.4 from the assigning authority and 8.8 from the national database, entered the US exploited-vulnerability catalogue on 12 August 2025 with a 2 September remediation date, and affects versions up to 7.12 [11]. Its ransomware-use field reads unknown while the aggregated metadata reads known, a conflict at source.
- Patch internet-facing web application firewalls, and treat a staged exploit as intent. Both teams recovered a public proof of concept for FortiWeb CVE-2025-25257 from the group’s infrastructure, alongside Linux-compatible tooling, without confirming in-the-wild use [1][2][3]. CERT-EU rates it CVSS 9.6, an unauthenticated SQL injection across the 7.0, 7.2, 7.4 and 7.6 branches [10].
- Detect on behaviour. Monitor unexpected PowerShell activity and payload downloads from public code-hosting services, new scheduled tasks, and writes to the Startup folder [1][2]. These three steps are common to all four documented infection chains, and none depends on which of nine names applies.
The infrastructure is the same story: publicly accessible staging directories, a previously unreported GitHub staging account, dynamic DNS domains, commercial VPN exits, and traffic blended through compromised third-party servers [1][3]. None of it is exotic, and most is rebuildable in an afternoon.
The operational point
This cluster has run for at least two years against ministries of foreign affairs, economy, energy, water and environment, using commodity delivery and lightweight interactive shells. The public record of it is nine names, one retracted implant, one reclassified family and a commit count that differs by one inside a single paper. That is not a criticism of the researchers, who did the consolidation work and published their corrections in the open. It is what actor-centric intelligence looks like as an operational input.
Take the indicators and the behavioural detections, which hold across every source, and treat the naming as metadata. The question for a government mail estate in this region is not whether UAT-6275 and TAG-193 are the same group. It is whether a shortcut file fetching a script from a public repository, creating a scheduled task and writing to the Startup folder would be seen tonight.
Sources
Every R3KONX article cites its primary material. 11 sources, in order of first citation. Links open the original publication.
- Break the silence: Kazakhstan-nexus UAT-6275 targets CIS government organizations Cisco Talos · 2026-10-14
- TAG-193's Regional Espionage Operations Recorded Future, Insikt Group · 2026-10-14
- Break the Silence: Tracking Silent Lynx Through Exposed Infrastructure Virus Bulletin Conference 2026, Seville · 2026-10-14 to 2026-10-16
- Silent Lynx: uncovering a cyber espionage campaign in Central Asia Virus Bulletin Conference 2025, Berlin · 2025-09-24 to 2025-09-26
- Silent Lynx APT Targeting Central Asian Entities Seqrite Labs · 2025-01-21
- Operation Peek-a-Baku: Silent Lynx APT and Dushanbe espionage Seqrite Labs · 2025-11-03
- ShadowSilk Group-IB · 2025-08-27
- ShadowSilk Campaign Targets Central Asian Governments Infosecurity Magazine · 2025-08-27
- Attributing YoroTrooper Cisco Talos · 2023-10-25
- Security Advisory 2025-024: Critical Vulnerability in FortiWeb (CVE-2025-25257) CERT-EU · 2025-07-11
- CVE-2025-8088 vulnerability record CIRCL Vulnerability-Lookup · published 2025-08-08, updated 2026-08-11
Researched and written by the R3KONX analysis desk from the cited primary material. Methodological caveat: this article compares vendor reporting that uses different names, different confidence language and different victim counts for overlapping activity; where the sources disagree, both figures are published rather than reconciled, and no attempt is made here to adjudicate which naming is correct. Attribution is reported only as the assessment of the named researchers. Corrections to event@r3konx.asia.
