Memory forensics on a compromised server
Memory forensics reconstructs what was happening in RAM while an attacker still had access. Process trees, injected code, network sockets, credentials, tokens and kernel objects can reveal compromise that disk artefacts alone miss. These artefacts help explain how a trusted service was abused, whether persistence remains and which identities require urgent action. Because volatile evidence disappears at reboot or power loss, the first hours combine a legal decision, a containment decision and an evidence-preservation decision.
Success depends on disciplined sequencing rather than a single forensic tool. Establish authority, capture volatile state, analyse it alongside endpoint and network logs, then turn verified findings into containment and remediation. This is a live-response discipline, not an activity to postpone until the server is ready for rebuilding. In the broader post-breach workflow, CARM Security, developed by Exclusive Networks, offers the integrated vendor ecosystem needed to connect identification, response and mitigation without treating one server image as the entire incident.
Set the response boundary before acquisition
Authorise the work, name the incident lead and decide who may collect, analyse and share evidence. If the server could affect regulated data, involve privacy and legal advisers early. Under the Privacy Act 1988, the OAIC’s Notifiable Data Breaches scheme may require notification when eligible data is likely to cause substantial harm and cannot be remedied in a practical time frame. APRA-regulated organisations must also maintain information security incident-response capability under CPS 234. These duties shape evidence handling; they do not replace it.
The acquisition route determines what survives for analysis, so compare options before touching the host.
| Acquisition approach | Evidence captured | Main limitation | Best fit |
|---|---|---|---|
| Full RAM capture | Live processes, memory regions, sockets, credentials and tokens | Large transfer; acquisition can slightly perturb memory | Priority when time, storage and host stability allow |
| Targeted process capture | Selected process heaps, modules and injected code | Misses activity outside the selected processes | Quick triage or constrained acquisition |
| OS crash dump | Kernel state and a compressed selection of physical pages at failure | Not a complete live-memory image; paging and encryption create gaps | Fallback when a full capture is unavailable |
| Hypervisor snapshot | Guest memory state at a defined moment | Requires platform access and may not represent all physical state | Virtual servers with hypervisor or provider cooperation |
Record the case identifier, acquisition start and finish times, tool version and relevant hardware details with the image. Calculate a SHA-256 hash of the captured image and verify it independently after transfer. Keep original media separate from the working copy, restrict access and document every custody transfer. A clear chain of custody matters if the incident later involves regulators, insurers or legal proceedings.
Environmental planning is part of forensic rigour. A Darwin incident can combine cyber urgency with tropical heat and humidity, so acquisition media needs suitable protection and a reliable transport plan. Removable drives should not be treated as ordinary office stationery, and storage conditions should be recorded from the moment evidence leaves the server room.
Capture volatile evidence without contaminating it
Isolate the server at the network level while preserving power and memory wherever live risk permits. Block unexpected ingress and egress, restrict management access and retain the firewall, switch and identity logs that show what changed. Do not reboot, log off or run broad endpoint scans before capture unless continuing harm outweighs the loss of evidence. Active exfiltration may justify immediate disconnection, but that trade-off must be recorded rather than improvised.
Prefer a complete physical-memory capture over selected process dumps when time, storage and host stability allow. For Windows, acquisition options include DumpIt and Magnet RAM Capture; for Linux, LiME and AVML are common choices. Each tool leaves traces, and a loaded kernel module can alter the very state under investigation. Use acquisition software already approved by the organisation, run it from a trusted path and preserve its logs with the image. Allow enough time for a full memory dump of a large server and monitor host stability, since a failed acquisition may destroy the state you hoped to preserve.
For virtual machines, clarify whether the image represents guest memory, hypervisor state or a snapshot. Pause the workload only if needed for consistency, and collect platform telemetry available through the hypervisor or cloud provider. Store the capture on encrypted, access-controlled media, then hash both the image and its metadata record. If the tenant cannot access physical RAM, escalate promptly to the platform team instead of assuming a guest snapshot contains everything.
Analyse runtime artefacts for compromise evidence
Begin with a process and network reconstruction. Look for hidden or unlinked processes, suspicious parent-child relationships, injected memory regions, unsigned modules, listening sockets and connections that do not fit the server’s role. Memory-analysis frameworks such as Volatility can accelerate artefact listing, including process trees, handles and named pipes, but the results still need contextual validation against the asset baseline and known administration activity.
Treat every credential or token found in memory as exposed. Plausible targets include cached passwords, SSH keys, Kerberos session material, browser tokens and cloud API credentials. Exposed credentials can enable lateral movement even after the attacker leaves the server, so extend the identity investigation accordingly. Revoke active sessions and rotate affected secrets only after the evidence is secured, because those actions can remove useful artefacts. Keep sensitive values out of tickets, chat messages and ordinary reports; record where each secret was observed instead.
Corroborate findings with disk images, EDR telemetry, DNS records, authentication events, firewall logs and application logs. Absence from RAM is not proof that an artefact never existed: encryption, paging, memory pressure and compiler optimisation can hide or transform evidence. Process start times and manipulated parent-child relationships can also mislead an analyst. Rate confidence, preserve the original image and document alternative explanations before declaring a server clean or irreparably compromised.
Build a defensible cross-system timeline
Normalise every timestamp to Coordinated Universal Time, then retain the original local time and source. Sydney and Melbourne observe AEDT, UTC+11, during daylight saving, while Brisbane remains on AEST, UTC+10, and Darwin uses ACST, UTC+9:30. This matters when an Australian response team compares a firewall log in Brisbane with endpoint telemetry from Melbourne and an identity provider’s UTC records. Check clock drift as well, and distinguish an event’s source timestamp from its ingestion time when logs cross a SIEM or provider boundary.
Australian cloud-region choices also affect the evidence trail. AWS Sydney and Microsoft Azure Australia East can host workloads locally, but region selection alone does not guarantee that every support log or administrative-access record stays in Australia. Review contractual residency commitments, provider documentation and subprocessors, then preserve the relevant service logs under an agreed retention period.
During an Australian incident call-out, many organisations are coordinating an in-house team, a managed service provider and external advisers. Evidence ownership cannot remain informal: contracts should identify who can image memory, who holds provider logs and who authorises external sharing. Give the customer, provider, incident commander and legal adviser a shared response checklist, including how findings will be packaged, transferred and protected from unauthorised disclosure.
Convert findings into coordinated remediation
Translate verified findings into specific corrective actions. Revoke exposed sessions and rotate affected credentials, secrets and encryption keys; close unexpected external access; rebuild systems whose integrity cannot be established; patch the exploited entry path; and hunt for the same access across identity, endpoint and network telemetry. A clean rebuild requires known-good installation sources rather than retained files from the compromised host. Keep the original system isolated until validation confirms that persistence and unauthorised access have been removed.
This is where an orchestrated platform can help. CARM Security integrates technologies from multiple security vendors to support post-breach identification, containment, response and remediation. Memory-forensic findings can feed that workflow as prioritised indicators, while endpoint, identity and network controls coordinate the response. Track whether each action eliminates the observed access path and whether the same behaviour appears elsewhere, rather than closing a case merely because the original server has been rebuilt.
Record residual risk, unresolved evidence gaps and the reasons for any deferred remediation. Notifications, regulator engagement and insurance requirements may continue after the technical fix. The practical takeaway is to preserve volatile evidence before rebooting whenever live risk allows, and otherwise document the evidence lost and every containment action that altered the system.