As defences for Cobalt Strike improved, intruders moved to other command and control (C2) frameworks that most detection stacks had less experience with: Sliver, Havoc and Brute Ratel C4. All three are offensive security tools built for legitimate red teams, and all three appear in criminal and state-linked intrusions according to MITRE ATT&CK, Microsoft, Unit 42 and Zscaler. Because each can be reconfigured, signatures for one build age quickly. This guide explains what each framework is, where it shows up, and how to detect and contain it through behaviours that do not change much between builds: delivery chains, process injection, credential access, lateral movement and network patterns. It is written for detection engineers, threat hunters and incident responders, and goes deeper than the overview in the Command and Control (C2) Guide. It does not describe how to operate any framework.
Key takeaways
- Hunt the behaviour, not the brand. Microsoft's Sliver research stresses detections built on technique, such as remote process injection and service-based lateral movement, so they survive a change of framework.
- The loader is often the easier catch. Sliver has been delivered by the Bumblebee loader, Brute Ratel by ISO and LNK packages with DLL search order hijacking, and Havoc by ClickFix lures and decoy documents.
- Commercial does not mean controlled. ATT&CK records that a cracked copy of Brute Ratel C4 leaked in September 2022, after which criminal use spread.
- Credential theft follows quickly. ATT&CK records LSASS memory access and Kerberos ticket abuse for Sliver, Kerberoasting support for Brute Ratel and token impersonation for Havoc.
- Network fingerprints are context, not verdicts. JARM, JA4, unusual HTTP header combinations and self-signed certificates impersonating Microsoft help prioritise, but every framework lets operators change them.
How the attack works
A C2 framework has three parts: a server the operator controls, listeners that receive implant traffic, and implants (Sliver "implants", Havoc "demons", Brute Ratel "badgers") that run on compromised hosts. The framework is post-exploitation tooling: it needs an initial foothold, usually from a phishing payload, a loader or a stolen remote access credential, and then provides command execution, file transfer, credential access and pivoting.
| Item | Sliver (S0633) | Havoc (S1229) | Brute Ratel C4 (S1063) |
|---|---|---|---|
| Type and licence | Open source, cross-platform, written in Go | Open source, first released on GitHub in October 2022 | Commercial red team product first appearing in December 2020; Unit 42 reports a licence of about $2,500 per user per year |
| Implant | Go binaries generated as executables, DLLs, services or shellcode; small stagers possible | Demon agent | Badger agent |
| C2 channels in ATT&CK | HTTP and HTTPS, DNS, mutual TLS, WireGuard | HTTP and HTTPS listeners, SMB named pipes between demons | HTTP and HTTPS, DNS over HTTPS, TCP, and web services including Slack, Discord and Microsoft Teams |
| Design focus noted by sources | Multiple operators, listeners and an extension package manager ("armory") | Wide post-exploitation feature set, community modules | Specifically designed to avoid EDR and antivirus detection |
| Abuse milestone | Microsoft observed adoption by threat actors since December 2020 | Adopted by multiple threat actors per ATT&CK | Cracked copy leaked September 2022 |
Brute Ratel's use of Slack, Discord and Teams as C2 channels is a form of living off trusted sites, covered in the Cloud Services Abuse for C2 Guide.
Where it shows up
| Framework | Observed use | Source |
|---|---|---|
| Sliver | Nation-state actors alongside or instead of Cobalt Strike; cybercrime groups supporting ransomware; the RaaS affiliate DEV-0237; human-operated ransomware delivered with Bumblebee (S1039) | Microsoft |
| Brute Ratel C4 | A May 2022 sample packaged as an ISO named as a CV, launching a signed Microsoft binary that side-loaded a malicious DLL; packaging resembled earlier APT29 (G0016) samples, but Unit 42 says packaging alone is not enough to attribute; Black Basta intrusions alongside QakBot and Cobalt Strike | Unit 42; ATT&CK (citing Trend Micro) |
| Havoc | A government organisation targeted in January 2023 through a ZIP with a decoy document and screen saver file; later ClickFix phishing campaigns and DLL side-loading lures | Zscaler; ATT&CK (citing Fortinet and Check Point) |
The same frameworks are used daily by authorised red teams, so a detection does not by itself mean a criminal intrusion. Confirm with the red team schedule, then treat any unexplained hit as an incident. See the Cobalt Strike Abuse Detection Guide for the framework these tools are replacing.
Impact
- Hands-on control: an operator can run commands, move files and take screenshots on any host with an implant (ATT&CK T1105, T1113).
- Credential compromise: LSASS access, token manipulation and Kerberos abuse turn one host into domain access (T1003.001, T1134, T1558.003).
- Lateral movement: SMB, WMI, WinRM and service creation spread implants (T1021.002, T1047, T1569.002); Sliver and Havoc pivot internally over SOCKS proxies and SMB pipes (T1090.001, T1559).
- Ransomware staging: Microsoft and ATT&CK tie these frameworks to ransomware affiliates, so a C2 implant is often the step before encryption. See the Ransomware Groups Guide.
Detection
Endpoint behaviours
| Behaviour | ATT&CK v20 | Telemetry | Notes |
|---|---|---|---|
| Signed binary launched from a mounted ISO or archive, loading a DLL from its own folder | T1574.001 DLL; T1204.002 Malicious File | EDR image load events; Sysmon Event ID 7 | Brute Ratel and Havoc delivery chains |
| Scripts or binaries running from user Temp after an archive or screen saver file is opened | T1204.002; T1059.001 PowerShell | EDR process trees | Havoc chain reported by Zscaler |
| Remote process injection: memory allocated, written and executed in another process | T1055; T1055.002 Portable Executable Injection | EDR injection alerts; Sysmon Event ID 8 (CreateRemoteThread) | Microsoft calls this a core Sliver signal |
| A process spawned and immediately injected into, especially a default sacrificial process | T1055 | EDR parent-child and injection correlation | Microsoft hunting guidance for Sliver |
| In-memory loading of DLLs without a file on disk | T1620 Reflective Code Loading | EDR memory scanning; writable and executable memory regions | Brute Ratel and Havoc |
| Access to LSASS memory by a non-security process | T1003.001 | EDR; Sysmon Event ID 10 | Sliver and post-exploitation generally |
| New services with random executable names in Windows Temp | T1543.003; T1569.002 | System log Event ID 7045; registry service keys | Microsoft notes Sliver's built-in lateral movement uses a service with default name and description |
| Changes to event tracing and in-memory security hooks | T1685 Disable or Modify Tools | EDR tamper and ETW telemetry gaps | ATT&CK records Brute Ratel patching ETW and AMSI |
Network behaviours
- Beaconing: regular check-ins with jitter to the same destination from one host; see the Beaconing and C2 Detection Guide.
- TLS context: self-signed certificates claiming to be Microsoft (Unit 42, Brute Ratel), mutual TLS from workstations to the internet, and JARM or JA4 fingerprints of listeners. Microsoft suggests probing suspected listeners for anomalous HTTP header combinations.
- Bare IP and new domains: Zscaler recommends alerting on HTTP or HTTPS to bare IP addresses from workstations and reviewing low-reputation or newly registered domains.
- DNS and DNS over HTTPS: high volumes of unique subdomain queries (Sliver DNS C2) and DNS over HTTPS from non-browser processes (Brute Ratel).
- Collaboration APIs from odd processes: Slack, Discord or Teams API calls from processes other than the official clients.
Response tip
Microsoft notes that Sliver's configuration must be decrypted in memory while the implant runs, so capture process memory before killing an implant if your forensics process allows it. That preserves C2 addresses for scoping and blocking.
Metrics
- Behaviour coverage: number of the endpoint behaviours above with a tested detection, validated by an authorised red team or emulation exercise.
- EDR in block mode: percentage of endpoints and servers with EDR in block mode and tamper protection.
- Injection alert triage time: median time from a remote injection or LSASS access alert to triage.
- Egress exceptions: number of servers allowed direct outbound internet access.
- Red team detection rate: share of authorised C2 framework exercises detected without a tip-off.
Prevention and hardening
- Break the delivery chain: block ISO, IMG, VHD and LNK attachments at the mail gateway, keep Mark of the Web enforcement (T1553.005 is the bypass to watch), and train users on ClickFix-style "paste this command" lures. See the ClickFix and Fake CAPTCHA Paste-and-Run Attacks Guide.
- Apply application control: allowlist what may run from user-writable paths and mounted images, starting in audit mode on a pilot group, then enforcing by ring with a documented rollback.
- Protect credentials: LSA protection and Credential Guard, gMSA for service accounts, and removal of RC4 for Kerberos.
- Control egress: servers should not reach the internet directly; workstations go through an authenticated proxy with TLS inspection where lawful, and DNS is forced through resolvers you log.
- Harden EDR: tamper protection on, block mode on, and alerts on loss of telemetry from a host, which is itself a signal of hook or ETW tampering.
- Manage your own red team tooling: register authorised framework infrastructure, certificates and test windows with the SOC so real intrusions are not dismissed as testing.
Where identity fits
C2 frameworks are a means to harvest and reuse identities: LSASS secrets, Kerberos tickets and impersonated tokens give the operator the accounts that matter. Containment therefore includes identity actions: reset or rotate credentials exposed on compromised hosts, revoke sessions, and look for new service accounts or tokens created during the intrusion. See the Service Account Security Guide.
Lessons from incidents
No live NHI Mgmt Group breach page documents one of these frameworks as the main tool, so the lessons come from the vendor research cited above.
- Undetected at first sight: Unit 42's Brute Ratel sample was flagged by no vendor when uploaded to VirusTotal, so behaviour detections were the only defence.
- Borrowed tradecraft confuses attribution: ISO packaging resembling APT29 did not make the activity APT29, a reminder to attribute on more than one artefact.
- Leaks change the threat: once a cracked Brute Ratel copy circulated, licence vetting no longer limited who used it.
- Loader-first detection works: Microsoft's observation of Sliver inside Bumblebee shows that catching the loader stops the framework before it runs.
Framework mapping
| Control | NIST CSF 2.0 | NIST SP 800-53 Rev. 5 | MITRE D3FEND |
|---|---|---|---|
| Application control and attachment blocking | PR.PS-05 | CM-7 Least Functionality | D3-EAL Executable Allowlisting |
| Endpoint behaviour monitoring | DE.CM-01 | SI-4 System Monitoring | D3-PSA Process Spawn Analysis; D3-PA Process Analysis |
| Egress filtering and network analysis | PR.IR-01 | SC-7 Boundary Protection | D3-OTF Outbound Traffic Filtering; D3-NTA Network Traffic Analysis |
| Credential protection | PR.AA-01 | IA-5 Authenticator Management |
Practitioner checklist
- Map your detections to the endpoint behaviours in this guide and fill gaps first for injection and LSASS access.
- Block ISO, IMG, VHD and LNK attachments and alert on signed binaries run from mounted images.
- Enable EDR block mode and tamper protection, and alert on telemetry loss.
- Alert on new services with random names in Windows Temp and on remote service creation.
- Log TLS fingerprints and flag self-signed certificates impersonating well-known vendors.
- Deny direct internet egress from servers and force DNS through logged resolvers.
- Capture process memory of live implants before eradication when possible.
- Register authorised red team infrastructure with the SOC and test detections against each framework.
- Rotate credentials and revoke sessions for every account used on a compromised host.
Standards and references
- MITRE ATT&CK: Sliver (S0633)
- MITRE ATT&CK: Havoc (S1229)
- MITRE ATT&CK: Brute Ratel C4 (S1063)
- Microsoft Security: Looking for the Sliver lining, hunting for emerging command and control frameworks
- Palo Alto Networks Unit 42: When pentest tools go brutal, red-teaming tool being abused by malicious actors
- Zscaler ThreatLabz: Havoc across the cyberspace
- MITRE ATT&CK: Process Injection (T1055)
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Command and Control (C2) Guide · Cobalt Strike Abuse Detection Guide · Beaconing and C2 Detection Guide · Cloud Services Abuse for C2 Guide