Skip to main content

Hit by ransomware? Isolate affected systems now. Do not reboot or reformat.

SheMo Noransom舍末无勒

Ransomware family

The Gentlemen Ransomware Decryption & Data Recovery

  • Active
  • Critical
  • Some versions decryptable

The Gentlemen is a RaaS operation that surfaced in mid-2025 and reached the top tier of global ransomware activity in 2026. It is marked by README-GENTLEMEN.txt notes and a six-character extension (.umc16h in the publicly analysed build), ships Windows and Linux/ESXi lockers, and combines self-propagation with an in-house EDR-killing framework.

First seen
2025-07
File extensions
.umc16h .[6位字符]
Ransom notes
README-GENTLEMEN.txt
Affected platforms
Windows / Linux / VMware ESXi

Family profile

File extensions
  • .umc16h
  • .[6位字符]
Ransom notes
  • README-GENTLEMEN.txt
Contact patterns
  • Tox ID printed directly in the Go-variant note (victim must install a qTox client)
  • A victim ID is printed alongside it, used to identify the victim at the negotiation entry point
  • Tor (.onion) leak site, where non-paying victims are published in stages
  • The C-based variant switches to email contact and drops the Tox ID
Aliases / versions
TheGentlemen、The Gentlemen RaaS、Storm-2697、GOLD SHERWOOD、ArmCorp(前身)
First seen
2025-07
Status
Active
Operational status
Actively operating
Threat level
Critical
Affected platforms
  • Windows
  • Linux
  • VMware ESXi
Tags
  • Leak-site regular
  • Active
  • Ransomware-as-a-Service
  • Double extortion
  • Targets virtualization
  • Exploits vulnerabilities
  • Phishing
Decryptor
Some versions decryptable

There is no general decryptor from law enforcement, No More Ransom or a mainstream vendor. Each file is encrypted with an ephemeral X25519 key pair and XChaCha20; the private key is never written to disk, so offline recovery is not possible in the normal case.

The only public recovery research is gentlemen-decryptor, published on GitHub by Bedrock Safeguard in 2026. The Go encryptor does not zero key material in heap memory after use, so if a process memory image happens to exist from the encryption window - an EDR/XDR auto-capture, a responder's procdump, a Windows Error Reporting dump, a full RAM forensic image, or a hibernation file written at the time - the ephemeral private keys can be extracted and matched against the public key in each encrypted file's footer.

The constraints are severe and must be stated plainly:

  • Without a memory image from the incident itself the tool is useless; capturing memory afterwards does not help.
  • Only files already encrypted before the memory snapshot can be recovered.
  • Files over 1 MB written in partial-encryption mode are skipped by default, to avoid corrupting untouched regions.
  • It comes from a single vendor with limited independent reproduction, so validate on offline copies before any bulk run.

For most engagements that begin after the fact this path does not apply, and recovery effort should centre on backups, snapshots and structural repair instead.

Sources

Latest activity

  1. Comparitech counted 675 attacks claimed on The Gentlemen's leak site since mid-2025, 600 of them in 2026. June 2026 was the first month it outposted Qilin (115 vs 78); July set a record 169.

    Sources
  2. Kaspersky disclosed a new C-based Gentlemen encryptor (AES-256-GCM + RSA, different ransom note) and a standalone ESXi locker, naming mainland China, Thailand, Indonesia, Brazil and Taiwan among its most-hit regions.

    Sources
  3. Check Point published The Gentlemen's leaked backend database, breached 4 May: internal chats, affiliate roles and tooling. One affiliate's C2 listed far more victims than the public leak site.

    Sources

Overview

The Gentlemen was first documented around July 2025. Unit 42 traces its core operators to ArmCorp, an affiliate crew of the Qilin RaaS; the programme opened public affiliate recruitment in September 2025 with a 90/10 revenue split against a market norm of 70-80%. Microsoft tracks the operators as Storm-2697, the Sophos counter threat unit as GOLD SHERWOOD.

Growth has been steep. Sophos counted fewer than 20 leak-site victims per month through late 2025, an average above 75 by early 2026 and a record 169 in July 2026, for 683 names between September 2025 and July 2026; Unit 42 separately recorded roughly 580 claimed victims across 77 countries as of early July 2026 and ranked the group the second most active RaaS programme behind Qilin. By sector, manufacturing was the largest slice of Unit 42's 580 (103 victims), while Sophos stresses a wide spread of industries and affiliates picking targets from whatever access they hold. Kaspersky telemetry places Brazil, China, Indonesia, Taiwan and Thailand among the more heavily hit regions.

On 4 May 2026 the administrator admitted that the group's own backend, "Rocket", had been compromised and internal data leaked; Check Point used it to publish the roles of roughly nine core members and their tooling. The crew kept operating afterwards. For enterprises in China the risk is not whether they are singled out, but how ordinary the entry points are: VPN without MFA, unpatched internet-facing firewalls, and directly reachable hypervisor management interfaces.

How to identify it

Extension. A six-character lowercase alphanumeric suffix is appended to the original filename - Sophos describes it as a "six-character extension", and the Go build Microsoft analysed used .umc16h. The base name is untouched, so order.xlsx becomes order.xlsx.umc16h. Public reporting does not establish whether the suffix is regenerated per build, so a different six-character suffix does not rule the family out - attribution has to come from the note.

Ransom note. README-GENTLEMEN.txt is dropped in every directory the encryptor finishes. The English text claims full control of the network, asserts that only the attackers' key can restore files, and gives a victim ID, a Tox ID and a publication deadline. The C-based variant Kaspersky disclosed in June 2026 uses a .txt filename that spells READ-ME and GEN-TLE-MEN in hyphenated segments (confirm the exact punctuation against the actual sample) and switches to email contact.

Other indicators. The Go encryptor is obfuscated with Garble and needs an 8-byte password argument to run, so executing the sample alone does nothing (sandbox evasion); the System, Application and Security event logs are cleared with wevtutil, shadow copies are deleted with vssadmin and wmic, and takeown, icacls and attrib reassign ownership and attributes. Domain environments show distribution via the NETLOGON share alongside PsExec and WMI execution; staged hosts commonly carry Advanced IP Scanner, Rclone, Restic, the SystemBC proxy and a Cloudflared tunnel installed as a Windows service.

Bottom line. The note plus a six-character extension attributes the family. Whether the build is Go or C and which encryption ratio was used comes only from analysing three to five real encrypted files and the original note - and that determines what recovery options remain.

Infection vectors

Sophos measured a median of roughly two days from initial compromise to encryption, with some incidents under 24 hours.

Initial access

  • Edge device vulnerabilities. Unit 42 records exploitation of CVE-2024-55591 in FortiOS/FortiProxy, CVE-2025-32433 in Erlang/OTP SSH, CVE-2025-33073 (Windows SMB client NTLM reflection) and CVE-2025-55182, plus suspected zero-day use.
  • Credential-based entry. VPN accounts without MFA - Sophos states plainly that "absence of MFA enabled the actor to gain access" - along with leaked or stolen credentials and access bought from initial access brokers.
  • Brute force and phishing. Credential brute forcing and spraying, and phishing delivery.

Internal spread. Microsoft found the Go encryptor carries its own self-propagation logic, attempting up to 21 lateral execution techniques per host - PsExec, WMIC, scheduled tasks, Windows services, PowerShell remoting, WMI - needing only one to succeed. Sophos additionally documents domain-wide distribution through NETLOGON shares, which is why domain-wide encryption overnight is the expected outcome.

Privilege and defence evasion. Mimikatz supplies credentials, and the in-house GentleKiller framework loads a vulnerable driver - Unit 42 names ThrottleStop.sys, CVE-2025-7771 - to reach kernel privilege and terminate security processes. Persistence relies on a custom Go backdoor, SystemBC, a Cloudflared tunnel installed as a Windows service, and scheduled tasks named gentlemen*.

Exfiltration and backup destruction. Rclone and Restic handle exfiltration. Before detonation the operators systematically disable Veeam, Backup Exec and SQL Writer services; Sophos catalogued over 200 command variations for this. Missing backups are usually not collateral damage from encryption - they are a deliberate prior step.

Encryption behavior

Algorithms. The Go encryptor generates an ephemeral Curve25519 (X25519) key pair per file, derives a shared secret against an embedded attacker public key, encrypts content with XChaCha20 and writes the ephemeral public key into the file footer. The private key is discarded after use and never touches disk, so there is no master key to attack offline. The C-based variant Kaspersky disclosed in June 2026 uses AES-256-GCM with RSA wrapping instead.

Partial encryption - the decisive factor in any recovery assessment. Files under 1 MB are encrypted in full. Larger files have chunks encrypted at three distributed offsets, at a ratio set by command-line flags: Microsoft puts the range at roughly 0.9% to 27%, and Kaspersky records --fast at about 9%, --superfast at 3% and --ultrafast at about 1%, with the default (around 27%) applying when no flag is passed. Large database files, virtual disks and mail stores therefore often retain substantial untouched data - but how much survives and how much can be repaired are different questions, turning on whether headers, page directories and partition structures fell inside an encrypted chunk.

Destructive actions. Around 30 processes (virtualisation, databases, backup software, EDR, SAP, Office, mail, browsers, remote access) are killed and some 50 services stopped to release file locks; takeown, icacls and attrib reassign ownership and attributes; shadow copies are deleted and event logs cleared.

Virtualisation and storage. The Go build Microsoft analysed targets Windows only; Kaspersky separately documents a C-based Linux/ESXi locker. In ESXi environments the encrypted objects are typically the flat disk files such as vmdk. Whether VMs were shut down first and which datastores were covered has to be established from on-site artefacts rather than assumed.

Double extortion. Data is stolen before encryption; non-payment leads to staged publication on the Tor leak site.

Assess before you act

Recoverability assessment

Recovery from The Gentlemen has to be assessed sample by sample and environment by environment. We do not pay ransoms and do not negotiate on a client's behalf - our work is technical recovery and forensics.

1) Public decryption tooling (very narrow conditions). No general decryptor exists from law enforcement or No More Ransom. The only published route depends on a process memory image preserved from the encryption window (an EDR auto-capture, a full RAM image or a hibernation file); if one exists, ephemeral private keys may be extractable. Without memory from the incident itself the path is closed, and capturing it afterwards does not help.

2) Repair space created by partial encryption (depends on the ratio). Files over 1 MB are encrypted only in chunks at three offsets, so MDF/LDF, DBF, ibd files, vmdk/vhdx disks and mail stores usually retain a large proportion of intact data; page-level extraction, logical rebuilds and virtual-disk structure repair are worth attempting. Yields depend on whether critical structures were hit and which ratio was configured. Assessment comes first; no conclusion is assumed.

3) Backups, snapshots and shadow copies. Shadow copies are normally gone and backup services usually disabled before detonation, but every option still needs checking: offline and offsite backups, read-only storage-layer snapshots (NAS and SAN snapshots frequently survive), hypervisor snapshots and cloud version history. This remains the highest-yield path in practice. Never reattach backup media to a network that may still hold implants.

4) Unencrypted copies and log replay. File-server recycle bins and version stores, endpoint caches, reporting staging databases, ERP/OA archive exports, database transaction logs and application audit logs can all support reconstruction or point-in-time replay.

5) Low-level carving. Some deployment modes write a new encrypted file and delete the original, leaving source data in unallocated clusters that raw sector scanning can recover - provided writes to affected volumes stop immediately.

Because exfiltration comes first, notification duties and credential rotation still apply even if every file is restored. We deliver a verifiable assessment and a clearly bounded recovery scope. We do not promise "100% decryption" or guaranteed restoration - given this family's cryptographic design, no technique supports such a claim.

Our response plan

Hit by The Gentlemen ransomware? What to do

  1. Isolation, memory preservation and forensic capture

    Cut affected hosts, ESXi servers and backup servers from production networks and storage paths - but do not reboot and do not power off. The only possible decryption path for this family depends entirely on process memory from the encryption window, which is lost permanently on restart, and implants such as SystemBC or a Cloudflared tunnel also need to be captured in memory.

    Order of work: collect memory first from hosts still encrypting or just stopped, then image or snapshot disks; next preserve the domain controller, backup server and hypervisor management hosts. Export logs from edge devices such as FortiGate, the VPN gateway, Active Directory and EDR - local event logs have been cleared, so rely on copies forwarded to a log platform. Keep three to five encrypted files and the original README-GENTLEMEN.txt.

  2. Family confirmation and encryptor build analysis

    Attribute from the ransom note rather than the extension, confirming this is The Gentlemen and not another family using a short random suffix. Then analyse the sample to answer the three questions that drive the recovery plan: Go build (X25519 + XChaCha20) or C build (AES-256-GCM + RSA); whether partial encryption was enabled; and which ratio was used - the ~27% default or fast, superfast, ultrafast.

    In parallel, check for artefacts of the driver GentleKiller loads, for NETLOGON distribution traces, and for which of the 21 lateral execution techniques were actually used, to establish whether the spread has been fully mapped. A missed host will re-encrypt everything after restoration.

  3. Parallel assessment of recoverability and breach impact

    Recovery side. Determine whether memory evidence from the encryption window exists; sample large files to measure the actual encrypted coverage and whether critical structures were hit; inventory offline backups, storage snapshots, hypervisor snapshots, cloud version history and untouched copies, and judge which media are trustworthy and which the attackers may have altered.

    Breach side. Use execution traces from Rclone and Restic together with outbound traffic records to scope which directories and what volume of data left, then assess exposure of personal information, trade secrets and contract data as the basis for notification duties and customer communication.

    The output is a per-system, prioritised feasibility statement: what can be recovered, the expected range, how long it will take, and what genuinely cannot be recovered.

  4. Recovery execution and service rebuild

    All work happens on copies; the original disks stay read-only. Sequence by business priority: rebuild the domain controller and identity systems first - in a clean environment rather than by restoring an old snapshot that may already contain an implant - then the ESXi cluster and core production VMs, then file services and endpoints.

    For databases, try trusted backups plus transaction log replay first; where no usable backup exists, move to page-level extraction and logical rebuilds. For virtual disks, repair partition and filesystem structures before extracting inner files. The recovery environment must stay isolated from the old network, every account credential is rotated before restoration, and reconnection is only allowed once the GentleKiller driver, SystemBC, the Cloudflared tunnel and gentlemen* scheduled tasks are confirmed removed.

  5. Root cause remediation and re-entry verification

    Identify the real entry point: check whether edge firewall management interfaces such as FortiGate were internet-reachable and unpatched against CVE-2024-55591, whether VPN access lacked MFA, whether account credentials are already circulating in leak or stealer data, and whether Erlang/OTP SSH or SMB/NTLM reflection attack surface existed.

    Hardening: remove public reachability from device management planes and patch them; enforce MFA on all remote access; constrain and monitor the lateral paths this family uses - PsExec, WMI, PowerShell remoting and scheduled tasks; alert on Group Policy and administrator group changes, suspicious registry modifications, new Defender exclusions and vulnerable driver loads; move backups to offline or immutable storage and treat backup service shutdown as a high-priority alert.

    Acceptance is measured against one question - could the attacker re-enter by the same route - and delivered with a repeatable verification report.

Risk warning

What not to do

  • Do not reboot or shut down hosts that are still encrypting or were just encrypted. The only possible decryption path depends on process memory from the encryption window, and a restart destroys that evidence permanently.
  • Do not install qTox to contact the operators or negotiate on your own. Payment neither stops already-exfiltrated data from being published nor prevents a repeat - it marks the organisation as one that pays.
  • Do not run unverified "decryptors" against original disks. Many such tools circulating online cause further structural damage; every attempt belongs on offline copies.
  • Do not rush to rebuild, reformat or keep writing to affected volumes. That overwrites original data still sitting in unallocated clusters and directly reduces what carving can recover.
  • Do not reconnect offline backup media or standby servers to an uncleaned network. SystemBC and a Cloudflared tunnel installed as a Windows service are often still live, and reconnecting hands over the last remaining copies.
  • Do not draw conclusions by searching the extension online. A six-character suffix carries very little signal and may not be identical across incidents; identifying by extension leads straight to the wrong recovery plan.

Emergency response

Data already encrypted? Stop and let an engineer look first

We do not pay ransoms and we do not negotiate with attackers. Engineers run a free assessment first, then propose a recovery plan and a firm quote.

Related scenarios

Related industries

Similar families

FAQ

The Gentlemen Frequently asked questions

  • My files got a six-character extension such as .umc16h and a README-GENTLEMEN.txt note. Can they be decrypted?

    Direct decryption should not be counted on. The Gentlemen generates a one-time X25519 key per file and encrypts with XChaCha20; the private key never touches disk, there is no master key to attack offline, and no general decryptor is published by law enforcement or No More Ransom.

    "Not decryptable" is not the same as "gone", though. Three things actually determine the recovery ratio: whether process memory from the encryption window was preserved, how much of each file over 1 MB was really encrypted (only chunks at three offsets), and how intact offline backups and storage snapshots are. Sample analysis answers that far better than searching the extension online.

  • I read there is a free decryptor for The Gentlemen. Does it work?

    One exists, but its preconditions are extremely narrow. Bedrock Safeguard published a tool in 2026 that exploits the Go encryptor's failure to zero key material in heap memory, extracting ephemeral private keys from a process memory image captured while encryption was running.

    That image has to already exist - an EDR auto-capture, a procdump taken at the time, a Windows Error Reporting file, a full RAM image, or a hibernation file written then. Capturing memory afterwards does not help, and a reboot removes it entirely. The tool also skips files over 1 MB written in partial-encryption mode, to avoid corrupting untouched regions. It comes from a single vendor with limited independent reproduction, so validate on offline copies first. For most incidents reported after the fact, this route does not apply.

  • Every VM on our ESXi host is encrypted. Is there anything left in the vmdk files?

    It is worth assessing. Files over 1 MB are only partially encrypted - chunks at three offsets - so a large share of the original data inside a vmdk is often untouched.

    In practice we test on copies where the encrypted chunks landed and whether the disk header, partition table and filesystem metadata were hit, then decide between structural repair followed by inner-file extraction, or pivoting to snapshots and backups. Read-only snapshots at the shared-storage layer (vSAN, SAN, NAS) must also be checked - in many cases they were never touched and deliver the best yield. No writes to the original datastore.

  • Our backup services were stopped and jobs failed. Was that caused by the encryption?

    No - it is normally a deliberate step taken before encryption. Sophos catalogued over 200 command variations this crew uses to disable Veeam, Backup Exec, SQL Writer and similar backup and database services, alongside deleting shadow copies and clearing event logs.

    Two consequences follow. First, a sudden run of failed backup jobs should be treated as an intrusion signal rather than an operations glitch - investigate instead of re-running the job. Second, offline and offsite backups, read-only storage snapshots, hypervisor snapshots and cloud version history still need to be checked one by one: what the attackers can stop is the online backup service, and genuinely offline or immutable media are often out of their reach. When verifying, never reattach backup media to an uncleaned network.

  • Why did the whole domain get encrypted overnight, and will it happen again after we restore?

    Because the encryptor propagates on its own. Microsoft's analysis shows it tries up to 21 lateral execution techniques per target host - PsExec, WMIC, scheduled tasks, Windows services, PowerShell remoting, WMI and others - needing only one to succeed, while Sophos additionally documents domain-wide distribution through NETLOGON shares. Sophos measured a median of about two days from intrusion to deployment, sometimes under 24 hours.

    Whether it recurs after restoration depends on two things. First, whether the entry point is genuinely closed: device management planes no longer internet-reachable, MFA enforced on VPN access, the relevant vulnerabilities patched. Second, whether persistence is fully removed: the GentleKiller driver, SystemBC, Cloudflared tunnels, gentlemen* scheduled tasks and any newly created domain administrator accounts. One uncleaned host or one un-rotated domain account is enough for the restored environment to be encrypted again, which is why root cause remediation and verification are a separate step in our process.