Ransomware family
Chaos Ransomware Decryption & Data Recovery
- Active
- High
- No public decryptor
Chaos is a cross-platform ransomware-as-a-service crew that surfaced in February 2025 and is assessed by Cisco Talos, with moderate confidence, to be a regrouping of former BlackSuit (Royal) members. It appends .chaos, drops README.chaos.txt, breaks in through Teams vishing, and layers DDoS threats on top of encryption and leak-site extortion.
- First seen
- 2025-02
- File extensions
- .chaos
- Ransom notes
- README.chaos.txt
- Affected platforms
- Windows / Linux / VMware ESXi / NAS storage
Family profile
- File extensions
- .chaos
- Ransom notes
- README.chaos.txt
- readme.chaos.txt
- Contact patterns
- Per-victim Tor (.onion) registration and negotiation portal; link and ID printed in the note
- Tor leak site, used to name victims and stage data releases
- A support mailbox in the note; Talos recorded samples using the thesecure[.]biz domain
- RAMP Russian-language crime forum, where the crew advertises the encryptor and recruits affiliates
- Aliases / versions
- Chaos RaaS、Chaos Ransomware Group(2025)、疑似 BlackSuit / Royal 后继团伙、STAC4749(Sophos 追踪的关联投放集群)
- First seen
- 2025-02
- Status
- Active
- Operational status
- Back in action
- Threat level
- High
- Affected platforms
- Windows
- Linux
- VMware ESXi
- NAS storage
- Tags
- Resurgent
- Active
- Rebranded
- Ransomware-as-a-Service
- Double extortion
- Targets virtualization
- Targets NAS
- Phishing
There is no public free decryptor for the Chaos RaaS encryptor. File contents are encrypted with AES-256 and each session key is protected by elliptic-curve key agreement (Talos documents Curve25519 ECDH). The private key stays with the operators, and no key leak or implementation flaw has been published, so the keys cannot be recovered offline.
A high-risk naming collision must be called out. No More Ransom does list a "Chaos Decryptor" contributed by Truesec, but neither the listing nor the tool repository states which build it covers; the same repository also ships decryptors for Onyx2 and SolidBit, variants descended from the 2021 Chaos builder lineage, which points at that low-grade commodity branch rather than the 2025 RaaS described here. Until a vendor confirms otherwise, do not assume it can process files carrying the .chaos extension - a mismatched tool that misreads the file format can damage the originals.
The correct first move is family and build identification: analyse three to five encrypted files, the original note and, where available, an encryptor sample to establish whether this is the RaaS encryptor or a same-named builder variant, and only then choose a recovery route. Any tool trial happens on offline copies, never on the original disks.
Latest activity
Sophos detailed STAC4749: Microsoft Teams vishing impersonating IT helpdesk hit dozens of North American organisations from February to June 2026 (Canada ~50%, US ~45%); at least three intrusions deployed Chaos, one encrypting within 17 hours.
SourcesRapid7 disclosed an intrusion that looked like Chaos but was assessed as Iranian-linked MuddyWater using the brand as a false flag: data theft with no encryption, Teams social engineering, credential harvesting and MFA tampering.
SourcesExtraHop reported Chaos now bundles DDoS as a standard affiliate capability, adding a third extortion layer beyond encryption and leaks; its encryptor uses -killvm via esxcli and a self-restoring -delete parameter.
Sources
Overview
Chaos was first observed in February 2025, advertising its cross-platform encryptor and recruiting affiliates on the Russian-language RAMP forum, and supplying builds for Windows, Linux, ESXi and NAS plus a victim-management panel - a textbook ransomware-as-a-service model. Cisco Talos assessed with moderate confidence that the crew is either a rebranding of BlackSuit (Royal) or operated by some of its former members, citing similarity in encryption parameters, note structure and tooling; the brand became busy after an international law-enforcement operation seized BlackSuit infrastructure on 24 July 2025. Vendors differ on the start date: Talos records first observation in February 2025 while others describe Chaos as forming only after that takedown; this page follows the Talos timeline.
The naming collision has to be settled first. This Chaos is not the 2021 Chaos ransomware builder behind Yashma and Chaos-C++; Talos notes the actors exploited the confusion around the name to obscure their identity. A leak site calling itself Trinity of Chaos is yet another distinct entity. The name alone identifies nothing.
Public leak-site aggregators list roughly 89 named organisations since February 2025, about 62 of them in the United States, with posts continuing into September 2026 - a count that moves over time. Sectors cluster in manufacturing, professional services, energy and construction. The RaaS rules claim the crew avoids BRICS and CIS countries, hospitals and government - a unilateral statement only. Rapid7 also documented, in May 2026, an operation borrowing the Chaos brand as a false flag, linked with moderate confidence to an Iranian state-aligned APT, which stole data without encrypting.
How to identify it
Extension and note. The .chaos suffix is appended to the original filename, so report.xlsx becomes report.xlsx.chaos. README.chaos.txt appears in affected directories, usually lowercase on Linux and ESXi. The English text frames the attack as a "security test", gives a victim-specific .onion registration and negotiation address, an ID and a support mailbox, and threatens publication, DDoS against internet-facing services, and spreading news of the breach to competitors and clients. Talos documented one demand of roughly $300,000.
Intrusion-side artefacts, which usually predate the note:
- Spam email flooding, followed by a phone call or an external Microsoft Teams chat and voice call from someone claiming to be IT support.
- Unexpected installs of remote assistance and RMM software: Quick Assist, RemSupp, AnyDesk, DWAgent, ScreenConnect.
- GoodSync used as a bulk exfiltration channel.
- Shadow copy deletion via vssadmin, Kerberoasting, Impacket executing commands over SMB, outbound reverse SSH tunnels.
A decisive branch. If a note exists but nothing is actually encrypted, treat the case as brand impersonation: hunt for long-dwell access and tampered MFA settings rather than starting a recovery workflow.
Infection vectors
Chaos barely relies on exploits; people are the entry point:
- Email flooding plus voice phishing. The target's mailbox is swamped with spam, then someone posing as the IT service desk calls or opens an external Microsoft Teams chat, offering to fix the mail problem and walking the user into a Quick Assist or RemSupp session. The STAC4749 cluster Sophos tracked hit dozens of organisations between February and June 2026, roughly 95% of them in North America, split about evenly between Canada and the United States.
- RMM persistence. The first session installs commercial remote-control and RMM software - AnyDesk, DWAgent, ScreenConnect - and enables RDP, so removing one tool does not cut access.
- Credential access and lateral movement. Kerberoasting for service account tickets, then Impacket executing commands over SMB, plus RDP and reverse SSH, with the domain controller as the objective.
- Exfiltration and deployment. GoodSync moves data out; the encryptor is then pushed over SMB and admin shares to endpoints, servers and hypervisors, detonating at once.
Speed is what makes this crew dangerous: in one Sophos case fewer than 17 hours separated the first Teams contact from estate-wide encryption, so controls have to sit on the vishing and remote-tool steps.
Encryption behavior
Algorithms. File contents are encrypted with AES-256 and each per-file key is protected by elliptic-curve key agreement; Talos documents Curve25519 ECDH. Key material is never written in cleartext and no exploitable flaw has been published. A few second-hand write-ups describe the key wrapping differently from Talos, so the reversed sample from the incident decides.
Configurable intermittent encryption. An /encrypt_step parameter sets what percentage of each file is encrypted, from 0 to 100, defaulting to 30. This is the single most important fact for a recoverability assessment: at the default, roughly seventy percent of a large file is untouched, but repairability depends on whether the overwritten regions include the header or the database page directory, so it must be measured.
Other runtime parameters. /lkey supplies the encryption key, /work_mode sets the scope (local, network or both), /ignorar_arquivos_grandes governs handling of very large files, and /kill_vms stops virtual machines before encryption. A recovered command line is direct evidence of the scope and intensity of this particular run.
Reach and recovery inhibition. The encryptor is multithreaded and enumerates and queues local disks alongside reachable network shares, so NAS devices and file servers are commonly encrypted in the same pass. It deletes volume shadow copies via vssadmin, and the ESXi/Linux builds stop virtual machines before encrypting vmdk and similar virtual disks.
Triple extortion. Non-payment leads to staged publication, DDoS against internet-facing services, and threats to spread news of the breach to clients and competitors. DDoS is stated explicitly in the note, so declining to negotiate has to be paired with availability protection, not only restore planning.
Assess before you act
Recoverability assessment
How much comes back depends on the encryption configuration, the backups and how the scene was handled. We do not pay ransoms and do not negotiate for clients; our work is technical recovery and forensics, and every assessment rests on measurement.
1) Free decryptor: none. Be careful with the similarly named Truesec "Chaos Decryptor" on No More Ransom: its listing states no supported build, and the repository shipping it also covers Onyx2 and SolidBit, pointing at the 2021 builder lineage. Until a vendor confirms otherwise, do not assume it applies to .chaos files - misuse can damage data.
2) Repair space from intermittent encryption (configuration dependent). At the default /encrypt_step of 30, large files retain substantial untouched regions: page-level extraction and logical rebuilds are worth attempting for database files, and repairing structures can let vmdk or vhdx images be mounted and extracted. But the parameter can be raised to 100, and critical structures may have been hit regardless, so sampling tests must precede any scope commitment.
3) Backups, snapshots and shadow copies. Shadow copies are usually gone, but a full inventory still pays: offline and offsite backups, immutable copies, storage-layer and hypervisor snapshots. Because Chaos encrypts reachable network shares in the same pass, online backups are often caught too. Never reattach backup media to an uncleaned network.
4) Unencrypted copies and carving. File-server recycle bins, BI staging databases, ERP archive exports and transaction logs often support reconstruction of critical records. If originals were deleted and rewritten, data may survive in unallocated clusters and can be carved by sector-level scanning, provided writes stopped immediately.
We never promise complete decryption, and no technique can assure that all data comes back; the achievable ratio follows from sampling tests.
Our response plan
Hit by Chaos ransomware? What to do
Containment and forensic preservation
Cut affected hosts, hypervisors and NAS devices off from networks and storage paths, and simultaneously block outbound ports for AnyDesk, DWAgent, ScreenConnect and similar tools - Chaos rarely leaves only one channel. Do not reboot or power off. Image memory and disk on the domain controller, backup server and any endpoint used as a jump point first, export Microsoft Teams external communication records, mail gateway, VPN and Active Directory logs, and preserve three to five encrypted files plus the original README.chaos.txt. Freeze the socially engineered user's account and session tokens at the same time.
Family identification and authenticity check
Confirm from the extension, note structure and sample characteristics that this is the Chaos RaaS encryptor rather than a same-named builder variant or another actor borrowing the brand. Two questions dominate: did encryption actually occur (if not, handle it as the state-sponsored false-flag scenario Rapid7 documented), and what /encrypt_step and /work_mode values were actually used. Reverse the encryption and key-wrapping routines and measure which regions of each file were overwritten - this decides whether the path forward is structural repair or backup rollback.
Recoverability and breach impact assessment
Run sample repairs against critical databases, virtual disks and file shares in an isolated environment while inventorying offline backups, immutable copies, storage and hypervisor snapshots and unencrypted duplicates. In parallel, scope the breach: reconstruct the volume, timing and content of exfiltration from GoodSync transfer logs and egress traffic, and determine whether regulatory notification is triggered. Deliver a written conclusion stating which systems go the structural-repair route, which rely on backup rollback and which are deferred, with expected recovery ranges, timelines and a business restoration order - then execute only after sign-off.
Recovery execution and return to service
All work happens on images or copies with originals kept read-only. Progress by business priority: rebuild a clean domain controller and identity layer first, then core databases such as ERP and MES, then virtual machines, file and mail systems. In ESXi cases, repair virtual disk structures and mount them to extract inner data rather than overwriting the original datastores. Because Chaos applies DDoS pressure, arrange traffic scrubbing and bandwidth protection before internet-facing services go back online. After each batch, run integrity checks and business-side verification - reconciliation, report comparison, application start-up tests - and record everything in a traceable recovery manifest.
Attribution, hardening and handover
Reconstruct the full kill chain: which external Teams account or phone number initiated the vishing, who approved the remote assistance session, when RMM tools landed, where domain administrator credentials leaked, and the timing and volume of exfiltration. Remove every remote-control tool, rogue account, scheduled task and PowerShell backdoor used for persistence, then reset credentials domain-wide and enforce MFA. Hardening must match the entry point: restrict external-domain Teams chat and calls, disable or tightly govern Quick Assist, application-allowlist RMM software, segment the ESXi management network and disable SSH, and rebuild backups to a 3-2-1 design with immutable copies. Close with an incident report, social-engineering awareness recommendations for staff and a formal handover checklist.
Risk warning
What not to do
- Do not reboot or power off affected hosts and ESXi servers - losing memory-resident key material, remote-control processes and live sessions destroys forensic evidence and any partial recovery chance at the same time.
- Do not download a "Chaos decryptor" from No More Ransom or elsewhere and run it against the original disks; those tools state no supported build and mostly target variants of the same-named 2021 builder, so a mismatch can cause further damage.
- Do not delete README.chaos.txt, encrypted samples or the remote-control installers, and do not rush to "clean the virus" - they are the only basis for family attribution, authenticity checks and exfiltration tracing.
- Do not contact the actors as the note instructs or pay the ransom; payment neither assures a working decryptor nor stops publication or DDoS pressure.
- Do not reattach backup tapes, external drives or the backup server to an uncleaned network - Chaos encrypts reachable network shares in the same pass.
- Do not declare the incident closed after removing a single remote-control tool without establishing the entry point; this crew typically leaves two or more remote channels in place.
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
ESXi / Hyper-V Virtualization Encrypted by Ransomware
Hypervisor-level encryption causes the widest blast radius of any ransomware event: dozens of production VMs go dark within an hour or two. This page covers what Linux ESXi encryptors actually do — shut down guests, encrypt vmdk, delete snapshots — the recovery value of flat disk files, and how Hyper-V and Proxmox cases differ.
File Servers and NAS Encrypted by Ransomware
When shared folders on a file server or NAS are encrypted, drawings, contracts, archives, quotations and design sources all become unusable at once — and mapped drives spread the impact to every endpoint. This page covers how to gauge spread, what shadow copies and snapshots realistically offer, and how to sequence recovery by business value.
Domain Controller Compromise and Estate-Wide Encryption
A compromised domain controller hands the attacker a legitimate administrator identity, allowing an encryptor to be pushed to every host at once through Group Policy or remote execution. This page covers how such incidents present, the correct order for Active Directory recovery, and how to decide between cleanup and full rebuild.
Backups Deleted or Destroyed
Modern ransomware follows a fixed sequence: destroy the backups, then encrypt the data — deleting shadow copies, encrypting repositories, disabling jobs, and exploiting backup software flaws to steal credentials. This page covers what can still be inventoried once backups fail, why replication propagates encrypted files off-site, and what offline and immutable copies are really worth.
Related industries
Manufacturing Ransomware Response and Recovery
Ransomware in manufacturing hits information systems and production cadence at the same time: with ERP down there are no orders, with MES down there is no schedule, and an encrypted drawing library takes the process documentation for an entire product line with it. This page covers the asset profile, recovery priorities and targeted defences.
Construction and Real Estate Ransomware Response
In construction and real estate the core assets are drawings, models and project documentation — typically scattered across project-site NAS units, shared drives and personal computers with no unified backup. This page covers the sector's threat profile, recovery of drawings and BIM models, and defences for dispersed, multi-project environments.
Healthcare Ransomware Response and Recovery
When a hospital is hit, registration, consultation, orders, billing, laboratory and imaging fail at the same moment and care falls back to paper. This page covers the healthcare threat picture, a recovery priority built around clinical continuity, and the handling of patient data and compliance obligations.
Similar families
- Some versions decryptable
Conti
Conti was one of the most destructive ransomware-as-a-service operations of 2020-2022, run by Wizard Spider (the TrickBot crew), with over 1,000 victims and more than USD 150 million collected. The brand dissolved in 2022 after its internal chats and source code leaked, and its members dispersed into Black Basta, Royal, Akira and other successors.
- Some versions decryptable
Akira
Akira is a ransomware-as-a-service operation that emerged in March 2023, breaking in through VPNs without MFA and edge-device flaws, then encrypting Windows estates and VMware ESXi clusters under double extortion. CISA's November 2025 advisory update calls it an imminent threat to critical infrastructure.
- No public decryptor
Interlock
Interlock emerged in September 2024 as a double-extortion crew marked by .interlock / .1nt3rlock extensions and the !__README__!.txt note. It relies on drive-by downloads, ClickFix fake-CAPTCHA social engineering and edge-device zero-days, and was covered by a CISA #StopRansomware advisory in 2025. No public decryptor exists.
FAQ
Chaos Frequently asked questions
Can files with the .chaos extension be decrypted?
Not with a generic tool. Chaos encrypts file contents with AES-256 and protects each session key through elliptic-curve key agreement; the private key stays with the attackers and no published flaw or key leak can be exploited.
"Cannot be decrypted" is not the same as "cannot be recovered". By default only about 30% of each file is encrypted, so large files such as databases, virtual disks and mail stores usually retain substantial intact regions that structural repair can exploit. Offline backups, immutable copies and storage or hypervisor snapshots often deliver the highest recovery ratio. The actual yield has to be established by sampling, never extrapolated from the percentage.
No More Ransom lists a Chaos decryptor - why does it not work on my files?
Most likely it is the same name for a different thing. Neither the No More Ransom listing nor the repository behind Truesec's "Chaos Decryptor" states which Chaos build it covers, and the same repository also ships decryptors for Onyx2 and SolidBit - both commodity samples descended from the 2021 Chaos ransomware builder. This page describes the Chaos RaaS crew that appeared in 2025, and Cisco Talos notes those actors exploited the security community's confusion over the name "Chaos" to obscure their identity.
The check is simple: if your files carry the .chaos extension and the note is README.chaos.txt pointing to a per-victim .onion registration and negotiation address, you are dealing with the RaaS lineage and there is no evidence that tool applies. Do not trial it on the original disks; a mismatched decryptor can destroy data that was still repairable. If you do need to try, ask the tool's authors to confirm the supported builds first.
A README.chaos.txt appeared - has our data already been stolen?
Assume exfiltration occurred, then confirm the scope with evidence. Chaos runs classic double extortion and routinely uses the legitimate sync product GoodSync to move data out before encryption, publishing it in stages if payment is refused and adding DDoS and calls to clients and competitors as further pressure.
What matters is what left, how much and when. Useful evidence includes GoodSync and similar transfer logs, egress statistics from firewalls and proxies, access auditing on cloud storage and file servers, and archiving artefacts on endpoints. This conclusion drives personal-information breach notification duties and customer disclosure, so it should be worked alongside legal counsel rather than taken from the ransom note's claims.
Every VM on our ESXi hosts is encrypted - is anything recoverable?
There is room to assess, but the scene must be preserved first. The Chaos Linux/ESXi encryptor carries a /kill_vms parameter that forces virtual machines down before vmdk files are encrypted, so one compromised host means every workload on it stops.
Three routes are worth testing. First, exploit the intact regions left by default intermittent encryption: repair the virtual disk partition table and filesystem structures, then mount and extract the inner data. Second, check hypervisor snapshots, storage-layer snapshots and untouched copies in the backup system - usually the highest-yield path. Third, for database VMs, page-level extraction and logical rebuilds often work even when the whole disk cannot be restored.
The precondition is that datastores are not reinitialised, RAID sets and storage pools are not rebuilt, and nothing writes repeatedly to the original volumes - otherwise low-level recovery options disappear for good.
Does Chaos attack organisations inside mainland China?
Public data shows Chaos victims are overwhelmingly in the United States and Canada, with no named mainland China victim in public leak-site tracking, and the crew's RaaS rules claim it avoids entities in BRICS or CIS countries. Monthly ransomware trend reports from Chinese vendors do not currently list Chaos among prevalent domestic families - as of September 2026 no Chinese-language analysis of the family was found, which is an absence of evidence rather than evidence of absence.
That is not a security boundary. First, such rules are unilateral claims, and its own leak site has carried healthcare victims it supposedly avoids. Second, North American and European subsidiaries, joint-venture plants and overseas sales entities of Chinese groups sit squarely in its target space, and those branches often share a domain, VPN and backup chain with headquarters. Third, Teams vishing plus remote assistance has no geographic barrier and works equally well against domestic teams. The practical priorities are restricting external Teams communication, governing Quick Assist and RMM tooling, and maintaining backups with immutable copies.
Sources
- Cisco Talos – Unmasking the new Chaos RaaS group attacks
- Sophos – Chaos in Teams vishing (STAC4749)
- ExtraHop – CHAOS in a BLACKSUIT: Triple Extortion Ransomware
- Rapid7 – Muddying the Tracks: The State-Sponsored Shadow Behind Chaos Ransomware
- BleepingComputer – Microsoft Teams vishing attacks lead to Chaos ransomware attacks
- No More Ransom – Decryption Tools(Truesec "Chaos Decryptor" 列表项)
- Truesec – TSDecryptors 工具仓库(Chaos / Onyx2 / SolidBit)
- Ransomware.live – Chaos 泄露站受害者统计
External links are provided for reference only. The content is published by third parties and does not represent our position.
Updated