Short · Network security · 17 min read
Can You Trust a Router or Firewall After a Factory Reset?
A factory reset can be part of recovering a compromised router or firewall, but its successful completion does not establish that the device is trustworthy. You need to know which state the procedure replaces, which state it preserves, how the replacement is authenticated and whether stolen credentials or compromised management systems can restore the attacker's access. The answer depends on the particular platform and incident, not the name of the reset button.
I would not automatically discard every suspect appliance. Nor would I reconnect one because its dashboard looks normal after a reset. The useful decision is whether the recovery procedure addresses the compromise you have evidence for, handles the important uncertainties and produces a state that can be accepted for the intended service.
That requires separating evidence preservation, containment, eradication and operational recovery. Those tasks can conflict in time, but treating them as the same task makes the conflict harder to manage. This article develops a proposed recovery assessment using twelve authoritative original sources and historical cases. Its method and fictional examples have not been empirically validated. No device was examined, reset, reimaged or connected to a network for this analysis.
Define what the recovery operation actually changes
Reboot, reset, upgrade, reinstallation and replacement are different operations. Their names are not standardized assurances that every storage area, identity or trust relationship has been renewed. Start with the vendor's documented behavior for the exact model, software train, deployment form and recovery mechanism. A virtual appliance and a physical appliance can require different procedures even when they carry the same product name.
I would map the device into relevant state classes: running memory, saved configuration, operating-system image, boot firmware, recovery material, additional storage or modules, local credentials and external management relationships. This is a proposed assessment inventory, not a claim that every appliance exposes each class separately. The point is to identify what the chosen action changes and what it leaves outside its scope.
NIST's firmware-resiliency guidance distinguishes protection, detection and recovery, with roots or chains of trust supporting those functions. It also treats firmware code and critical configuration data as distinct recovery concerns (Regenscheid, 2018, §§4.1, 4.4). Those distinctions help explain why resetting settings cannot automatically answer a question about the code that interprets them.
For each relevant state class, ask four questions: could the suspected access have changed it; does the recovery procedure replace or validate it; what supports the integrity of that procedure; and what evidence will remain afterward? An unknown answer should stay visible. Do not silently convert an undocumented reset behavior into an assurance that the state must have been erased.
The resulting scope might be narrow and sufficient. For example, a verified unauthorized configuration change could be remediated through an appropriately trusted platform and reviewed configuration process. A supported indication of altered boot components presents a different problem. The same operational symptom does not mean the same compromise depth, and the response should not assume that it does.
Preserve the evidence that the reset would destroy
A reset is an intervention in the evidence. Cisco's IOS forensic guidance warns that rebooting during initial triage loses volatile information, including memory and operational tables, and recommends isolation before examination to reduce adversary interference (Cisco, 2026b, Introduction). That is platform-specific response guidance, not a universal instruction to keep a dangerous device connected while collecting everything.
Before changing the device, decide what the investigation needs and what the response can safely preserve. A current connection table, running configuration, log buffer and memory artifact may answer different questions. The existence of an old configuration backup does not replace a record of what the appliance was doing when compromise was suspected.
Collection itself changes state. NIST's forensic guide explains both the value of volatile data and the risk that live collection alters evidence or receives misleading results from a compromised system (Kent et al., 2006, §5.2.1). I would use that principle to document actions and limitations rather than promise an untouched acquisition from a live network appliance.
Even a procedure described as collecting memory can be disruptive. Cisco's ASA guide warns that its core-file generation procedure alters configuration and triggers an immediate reload. It also limits its applicability to ASA software, excluding other operating systems and installed service modules (Cisco, 2026a, Introduction and “ASA Core File Generation”). Copying a familiar command sequence across products can therefore destroy evidence or create an outage without producing the artifact you expected.
The response lead should approve a sequence that reflects the incident's urgency and operational consequences. If immediate interruption is necessary, record why evidence could not be collected first and what was lost. If bounded collection is feasible, preserve the relevant outputs through an approved channel before destructive recovery. Neither choice should be disguised afterward as a complete forensic acquisition.
Contain the device without assuming it enforces its own containment
A suspect firewall is an awkward place to rely exclusively on a new firewall rule. The problem is not that every local command must be false; it is that the component whose integrity is questioned may also be producing the enforcement claim. Determine which restrictions can be imposed or observed through a separately trusted boundary.
That might involve an upstream control, physical disconnection of a defined path or a controlled replacement service. The appropriate mechanism depends on the architecture and the effects on essential operations. Identify the administration path that remains available after the restriction. A plan that severs the only recovery channel can leave the organization improvising a broad exception under pressure.
High availability needs particular scrutiny. Before failing over, establish what state the peer shares and how it is administered. A standby can be a recovery option, but the words standby and healthy do not establish independent trust. If it has received the same configuration or is controlled through the same compromised management system, simply moving traffic may preserve the exposure that prompted the response.
The broader containment-evidence analysis addresses how to substantiate a bounded restriction. Here the additional concern is self-reference: can the recovery team establish the boundary without relying solely on the device under investigation? Preserve external observations where possible, while explaining their scope and any dependency on the suspect component.
Operational continuity does not disappear from the decision. Where a network device supports safety-relevant processes, the isolation action belongs in the established operational and engineering authority structure. This paper supplies no shutdown sequence. The useful preparation is a defined alternative service and an approved way to reach the device that does not require reopening every original connection.
A signed image answers a narrower question than a clean device
An image's signature or expected hash can be valuable evidence about that image. It does not automatically establish the integrity of every component that loads it, every active configuration item or the environment that manages it. Identify precisely what was checked, by which mechanism and against which trusted reference.
NIST recommends that firmware recovery mechanisms and authentic recovery material be protected independently of the running firmware, and that recovery images be authenticated before restoration. Its guidance also addresses preventing unauthorized recovery to an older vulnerable version (Regenscheid, 2018, §4.4.1). These are architectural requirements to investigate, not evidence that any appliance you own implements them correctly.
I would distinguish authenticity from suitability. An authentic old image may still be inappropriate for the incident because it lacks the needed fix or conflicts with the supported recovery path. A newer image downloaded through an untrusted channel creates another problem. The selected artifact needs an explainable source, supported target and verification process; a version number alone does not supply all three.
The verifier also has a scope. If all results come from software that may have been subverted, repeating the same check several times does not create independent assurance. Ask the vendor which lower-level or external mechanisms can validate the relevant components and what cannot be assessed through ordinary administration. Escalate a genuine verification gap rather than inventing a command that supposedly proves the entire device clean.
Replacement may be justified where the required trust cannot be re-established through a supported procedure. That is different from declaring hardware replacement mandatory after every suspicious log entry. The decision should identify the unresolved state, the consequence of accepting it and the alternative that removes the uncertainty. Otherwise, a costly replacement can become another ritual rather than a reasoned recovery choice.
Read firmware incidents with their platform limits intact
The 2023 joint BlackTech advisory documented modified router firmware and, in some cases, modified bootloaders used for persistence across reboots. It also described adversary activity that interfered with ordinary visibility (National Security Agency et al., 2023, “Modified bootloader” and “Modified firmware”). The narrow lesson is that restarting a device does not necessarily remove compromise held outside volatile execution state.
That historical report should not be stretched into a claim that every modern router's secure boot is ineffective. Cisco's response stated that the reported activity involved compromised administrative credentials, that it had no indication Cisco vulnerabilities were exploited in those attacks, and that modern secure-boot capabilities prevent the loading and execution of modified images within their supported design (Cisco, 2023, Summary).
These sources establish different parts of the assessment. The advisory describes observed techniques and exposure; the vendor response narrows the platform and initial-access interpretation. Neither substitutes for examining your device's model, boot protections, software support and evidence of administrative compromise. A recovery review should retain both statements instead of selecting whichever produces the most dramatic headline.
I would ask which premise of a trusted startup chain remains supported after the incident. If administrative credentials were exposed, the recovery must address their continuing authority. If a legacy platform lacks the relevant protection, installing another copy of software may leave an architectural weakness. If evidence suggests a modified component beyond the normal reinstall scope, the plan must address that component explicitly.
Do not confuse absence of a published exploit with absence of harmful administrative capability. An attacker using a valid privileged account can make damaging changes without exploiting a new software vulnerability. Conversely, a firmware incident on another platform is not evidence that your own appliance suffered the same modification. The response needs specific findings and carefully bounded hypotheses.
The Ivanti case shows why lab results and field findings must stay separate
The February 2024 joint Ivanti advisory included CISA research on an isolated virtual appliance, identifying the tested version and laboratory conditions. Its appendix reported a persistence mechanism that survived resets and upgrades and could deceive integrity checks in that setting (Cybersecurity and Infrastructure Security Agency [CISA] et al., 2024, Appendix A). That is consequential evidence about a tested recovery limitation, but its provenance matters.
Mandiant's contemporaneous investigation reported attempts to persist across upgrades, patches and resets. It stated that the limited persistence attempts it observed had not succeeded because the malware did not account for an encryption-key mismatch; the report also described post-reset analysis finding no persistence on an examined appliance (Lin et al., 2024). An observed attempt and an observed successful persistence event are not interchangeable claims.
Ivanti's response disputed the applicability of the lab technique to live customer environments and distinguished hardware factory reset from deployment of a new virtual build. It recommended the updated external integrity checker alongside continued monitoring (Ivanti, 2024). Those were the vendor's stated conclusions and recommendations at that time, not independent proof that every reset was sufficient.
I would preserve all three positions in an evidence record: who tested what, on which version, in which environment, and what outcome was actually observed. The disagreement cannot be resolved by turning the most cautious statement into a universal fact or treating the vendor's assurance as a complete local verification. Nor should a 2024 procedure be applied unchanged to a different vulnerability or current appliance release.
The operational lesson is to obtain incident-specific recovery guidance and identify its assumptions. A reset may replace state differently on a virtual platform; a persistence attempt may fail under one key transition and succeed under another condition. Where that distinction affects acceptance, ask for a supported explanation of the recovery behavior and retain the versioned evidence. Historical cases should improve the questions, not become timeless commands.
Recover configuration as an authorized state, not just a backup file
A saved configuration may contain legitimate service requirements, accidental weaknesses and adversary changes at the same time. Restoring it wholesale can make the device operational while restoring the access path that enabled the incident. Treat the backup as a candidate input with provenance and a capture time, not an automatic source of truth.
NIST's configuration-management guidance distinguishes establishing secure configurations from controlling changes through security-impact analysis, testing and approval (Johnson et al., 2011, §§3.2.1, 3.3.2). Applied to recovery, this means that a pre-incident baseline and a post-incident export have different roles. The latter can preserve evidence and reveal dependencies without being approved for production reuse.
Review relevant differences in administrator identities, management reachability, forwarding policy, remote access, logging destinations and automatically loaded features. The exact list depends on the platform. A setting that was necessary before the incident may need replacement if it depends on compromised credentials or a management service whose integrity remains unresolved.
Manual reconstruction is not inherently safer. It can omit an essential restriction or introduce a rushed exception. I would use a reviewed target configuration, a recorded comparison and functional checks appropriate to the device. Where the vendor supports a selective import, document what was included and excluded; do not call a partial import complete merely because the main application becomes reachable.
A configuration difference also needs interpretation. Not every deviation from an old baseline is malicious, and not every unchanged setting is safe. Tie the recovery decision to the approved service and current threat finding. Preserve unresolved differences as open acceptance items rather than updating the baseline solely to remove the discrepancy from the report.
Device recovery does not revoke authority stolen from it
A reset changes the device; it cannot retrieve a secret that an adversary has already copied elsewhere. The 2024 Ivanti advisory explicitly told defenders to treat stored user and service credentials on affected appliances as likely compromised and to hunt for activity beyond the gateway (CISA et al., 2024, Summary). Its scope is that incident, but it illustrates why a device-only recovery boundary can be too narrow.
Map the identities and trust relationships the appliance could use or expose. These may include local administration, directory integration, management automation, VPN-related credentials or stored keys. Identify which were present in the actual environment and which exposure hypotheses the evidence supports. Do not declare every enterprise secret compromised solely because one device is suspect.
Replacement and revocation have to be coordinated with consumers. If an automation service immediately reconnects using an old credential, it may fail or restore an insecure fallback. If a secret was rotated while the compromised system could still observe the new value, the rotation may not have achieved its intended purpose. Sequence identity recovery with the establishment of a trustworthy administration path.
Separate changing a value from ending its authority. The old credential's acceptance must be addressed at the system that validates it; local deletion of a stored copy is not enough. Likewise, existing sessions, delegated permissions or downstream changes may require their own incident-specific action. A successful login with the replacement credential proves only that the replacement works.
This does not require waiting for perfect enterprise-wide certainty before restoring any service. It requires a clearly bounded trust decision: which identities were remediated, which dependencies were checked and what unresolved exposure prevents normal access. The network device can be technically rebuilt while the surrounding identity recovery remains incomplete.
Reconnect in stages that can reveal a wrong recovery assumption
A recovered device should first operate within a controlled scope appropriate to validation. Identify who can administer it, which peers it can reach and where its evidence is collected. If the same suspect management system immediately pushes configuration back onto it, the recovery has inherited a trust dependency that a fresh image alone does not resolve.
The 2025 edge-device practitioner guidance recommends segregated management interfaces and centrally protected logs. It specifically calls for controls independent of the edge device to limit its access to the log collection point, and for preventing devices from modifying or deleting stored logs (Australian Signals Directorate's Australian Cyber Security Centre [ASD's ACSC], 2025, “Secure management interfaces” and “Centralise monitoring for threat detection”). That provides an architectural basis for observation that the recovered device does not wholly control.
Begin with the operations needed to assess the recovery, then admit the required service relationships in an approved order. For each stage, record both the allowed operation and a relevant restriction. A firewall passing production traffic is not enough if management access is unintentionally exposed; a device denying everything is not a useful demonstration of restored service.
Define rollback or stop conditions before expanding connectivity. An unexplained configuration change, failed integrity check or unexpected use of a retired credential should trigger an explicit reassessment. The response need not prove the exact attacker mechanism before restricting a newly suspicious path, but it should distinguish a precautionary restriction from a confirmed recurrence.
Observation after reconnection matters, yet a quiet interval has limited meaning. It cannot prove that every persistence mechanism is absent or that the incident's original cause was removed. Combine monitoring with the recovery evidence already established, and report the interval and visibility limitations. Time passing is not a substitute for an unperformed verification step.
Decide acceptance through explicit recovery claims
NIST's current incident-response guidance calls for checking restoration assets before use, verifying restored assets and addressing root causes before production return. It also distinguishes the completion of recovery through defined criteria (Nelson et al., 2025, RC.RP-03, RC.RP-05–06). I would turn those expectations into a device-specific acceptance record rather than a generic clean-device certificate.
| Recovery claim | Evidence to review | Reason to withhold the claim |
|---|---|---|
| The relevant state was addressed | Model-specific reset/rebuild scope matched to findings | Suspected persistence lies outside documented scope |
| Recovery inputs were trustworthy | Authenticated supported images and reviewed configuration | Reference, delivery path or backup provenance unresolved |
| The recovery mechanism was suitable | Supported boot/recovery procedure and applicable verification | Verifier depends entirely on a suspect component |
| External authority was remediated | Defined credential,session and management-dependency actions | Old authority remains accepted or new material was re-exposed |
| Required service and restrictions work | Bounded positive and negative operational checks | Availability returned by reopening an unjustified path |
| Continuing observation is usable | Independent logging and explicit monitoring limitations | Evidence can be erased or is unavailable for the relevant activity |
This matrix is a proposed reasoning aid, not a standard certification. A failed row does not prescribe one remedy in every environment. It identifies why normal acceptance is not yet supported. The next action could be further vendor-assisted examination, a different rebuild path, restricted service or replacement, depending on the unresolved claim and business consequence.
Consider fictional Westhaven, whose gateway shows an unexplained administrative change. In the first constructed case, investigation supports a configuration-level incident, the approved recovery process validates the relevant software state, configuration is rebuilt from reviewed inputs, and exposed authority is revoked. A bounded return-to-service decision may be supportable under those facts.
In a second case, the team cannot explain whether the suspected modified component is replaced by the reset procedure. Repeating that reset does not remove the knowledge gap; the plan needs a supported deeper recovery or replacement. In a third, the device is rebuilt but its old privileged integration remains usable elsewhere. Hardware recovery has progressed while incident recovery has not finished.
These are invented cases, not client outcomes. Their purpose is to make the acceptance criteria discriminate between materially different conditions. The wider service-restoration analysis addresses business recovery performance and dependencies; this paper's narrower concern is whether trust in a particular network device has been re-established well enough for its assigned role.
When replacement is the better decision
Replacement becomes more compelling when the platform is unsupported, a required recovery mechanism is unavailable, relevant integrity cannot be assessed or repeated recovery efforts cannot establish an acceptable state. The edge-device guidance recommends retiring unsupported devices rather than treating compensating controls as a permanent substitute for support (ASD's ACSC, 2025, “Manage end-of-life devices”). That is an important lifecycle constraint, not evidence that a specific device is already compromised.
A replacement still needs authenticated software, reviewed configuration, new or appropriately remediated credentials and a trustworthy management relationship. Importing the suspect environment wholesale can reproduce its weakness on different hardware. Include the old device's evidence preservation and disposition in the plan so that replacing service does not accidentally destroy the remaining investigation material.
Cost should include the consequences of uncertain acceptance, not just the purchase price. Equally, an unbounded fear of firmware compromise can waste resources where evidence supports a narrower repair. I would compare supported recovery options against the actual unresolved claims, service deadline and available verification. There is no defensible universal rule that every incident requires replacement or that a reset always makes replacement unnecessary.
This analysis relies on technical guidance and scoped historical reporting, not comparative testing of appliance vendors. Recovery behavior, support status and incident-specific advisories can change. The 2023 and 2024 cases are not claims about every current product version. No single scan, signature check or observation interval can guarantee the absence of every possible compromise.
The decision I would want to sign is precise: this device, in this recovered state, with these remediated relationships, meets these service and security conditions; these uncertainties remain, and these findings would reopen acceptance. That is something an incident team can defend. “We reset it” is merely a description of an action.
Request a network-device recovery review to examine the scope of your reset or rebuild procedure, the trustworthiness of its inputs and the evidence required before returning a router, firewall or VPN gateway to service.
References
- Australian Signals Directorate's Australian Cyber Security Centre. (2025, February 4). Mitigation strategies for edge devices: Practitioner guidance. https://www.cyber.gov.au/publication/mitigation-strategies-for-edge-devices-practitioner-guidance
- Cisco. (2023, September 27). Reports about cyber actors hiding in router firmware. https://sec.cloudapps.cisco.com/security/center/content/CiscoSecurityAdvisory/cisco-sa-csa-cyber-report-sept-2023
- Cisco. (2026a, June 4). Cisco ASA forensic data collection procedures (Version 1.6). https://sec.cloudapps.cisco.com/security/center/resources/forensic_guides/asa_forensic_investigation.html
- Cisco. (2026b, February 2). Cisco IOS software forensic data collection procedures (Version 1.6). https://sec.cloudapps.cisco.com/security/center/resources/forensic_guides/ios_forensic_investigation.html
- Cybersecurity and Infrastructure Security Agency, Federal Bureau of Investigation, Multi-State Information Sharing & Analysis Center, Australian Signals Directorate's Australian Cyber Security Centre, United Kingdom National Cyber Security Centre, Canadian Centre for Cyber Security, New Zealand National Cyber Security Centre, & CERT New Zealand. (2024, February 29). Threat actors exploit multiple vulnerabilities in Ivanti Connect Secure and Policy Secure gateways (AA24-060B). https://www.cisa.gov/sites/default/files/2024-02/AA24-060B-Threat-Actors-Exploit-Multiple-Vulnerabilities-in-Ivanti-Connect-Secure-and-Policy-Secure-Gateways_0.pdf
- Ivanti. (2024, February 29). Enhanced external integrity checking tool to provide additional visibility and protection for customers against evolving threat actor techniques in relation to previously disclosed vulnerabilities. https://www.ivanti.com/blog/enhanced-external-integrity-checking-tool-to-provide-additional-visibility-and-protection-for-customers-against-evolving-threat-actor-techniques-in-relation-to-previously-disclosed-vulnerabilities
- Johnson, A., Dempsey, K., Ross, R., Gupta, S., & Bailey, D. (2011). Guide for security-focused configuration management of information systems (NIST SP 800-128; updated October 10, 2019). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-128
- Kent, K., Chevalier, S., Grance, T., & Dang, H. (2006). Guide to integrating forensic techniques into incident response (NIST SP 800-86). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-86
- Lin, M., Wallace, R., Larsen, A., Gandrud, R., Thompson, J., Pearson, A., & Frazer, A. (2024, February 27). Cutting edge, part 3: Investigating Ivanti Connect Secure VPN exploitation and persistence attempts. Google Cloud. https://cloud.google.com/blog/topics/threat-intelligence/investigating-ivanti-exploitation-persistence
- National Security Agency, Federal Bureau of Investigation, Cybersecurity and Infrastructure Security Agency, Japan National Police Agency, & Japan National Center of Incident Readiness and Strategy for Cybersecurity. (2023, September 27). People's Republic of China-linked cyber actors hide in router firmware (AA23-270A). https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-270a
- Nelson, A., Rekhi, S., Souppaya, M., & Scarfone, K. (2025). Incident response recommendations and considerations for cybersecurity risk management: A CSF 2.0 community profile (NIST SP 800-61 Rev. 3). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-61r3
- Regenscheid, A. (2018). Platform firmware resiliency guidelines (NIST SP 800-193). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.SP.800-193