After a ransomware attack, the team rebuilds a server: a clean system install, data restored from backup, new passwords. Two weeks later, suspicious traffic appears on the same server again. The system is fresh, EDR is silent, and yet something survived. If the attacker previously had access to the server’s management interface or changed its firmware, reinstalling the system removed nothing that mattered.
Operating system security assumes that the layer underneath is trustworthy. In this article we test that assumption: what a BMC controller is, how attackers use management interfaces, what a hardware root of trust provides, and what to ask when buying an x86 server. Lenovo ThinkSystem servers serve as an example of a concrete implementation, but the questions apply to every manufacturer.
Why is an attack on server firmware more dangerous than an operating system infection?
Because it works below everything that normally protects the server. A change in UEFI, in the management controller’s firmware or in motherboard chips survives a system reinstall and a disk replacement. Tools running inside the operating system, such as EDR or antivirus, do not see it, because they start later.
In SP 800-193 “Platform Firmware Resiliency Guidelines” (May 2018), NIST defines the platform as the basic hardware and firmware that a system needs in order to start and run. The publication notes that a successful attack on platform firmware can render the system inoperable, perhaps permanently, or require it to be reprogrammed by the manufacturer.
Malware that hides at this level is called a bootkit or firmware rootkit — we explain the terms in our glossary under rootkit. For defence this means one thing: control has to start in the hardware, not in the system it is meant to protect.
What is a BMC, and why is it a separate computer inside your server?
A baseboard management controller is a separate processor on the motherboard, with its own system, its own network interface and its own accounts. It runs whenever the server is connected to power — including when the operating system is off. It is used for remote administration: powering on and off, viewing the console, mounting media images, monitoring hardware and updating firmware.
Each manufacturer has its own implementation. Lenovo calls it XClarity Controller, HPE — iLO, Dell — iDRAC. According to the Lenovo Press product guide for the SR650 V4, XClarity Controller 3 is based on the ASPEED AST2600 BMC chip and supports OpenBMC. It provides standard management protocols: IPMI 2.0, SNMPv3 and the Redfish REST API.
From a security standpoint, the BMC has more privileges than the system administrator. Whoever takes over the controller can reboot the server from their own image, watch the console and change firmware. Backups, disk encryption and system passwords then lose their meaning, because the attacker operates beneath them.
The BMC in three sentences: it is a second computer inside the server, with its own network and accounts. It runs when the system is switched off. Whoever controls it controls the server — regardless of what is installed on the disks.
How do attackers exploit server management interfaces?
Most often by the simplest route: network access to a controller that should not be reachable. A BMC interface plugged into a flat office network or, worse, reachable from the internet becomes a scanning target. Default or weak passwords and outdated controller firmware do the rest.
The second route is vulnerabilities in the BMC software itself. An example from the CISA Known Exploited Vulnerabilities catalogue is CVE-2024-54085. The vulnerability in AMI MegaRAC SPx firmware, used by many server board manufacturers, allows authentication to be bypassed remotely through the Redfish Host Interface. CISA added it to its catalogue of actively exploited vulnerabilities on 25 June 2025.
This is an example of a threat class, not of a specific manufacturer discussed here. Whether a given vulnerability affects your servers is decided solely by their manufacturer’s security advisory. The general conclusion is simple, though: BMC firmware needs patching just like the operating system, and its vulnerabilities end up in catalogues of actively exploited flaws.
An example from KEV: CVE-2024-54085 concerns AMI MegaRAC SPx, not Lenovo XClarity Controller. It shows that BMC vulnerabilities are actively exploited — and that the manufacturer’s advisory decides which servers are affected.
What do a hardware root of trust and NIST SP 800-193 compliance mean?
It is an independent hardware element that is trusted unconditionally and checks everything that runs after it. NIST SP 800-193 bases firmware resiliency on three principles: protection against unauthorised change, detection of a change that has already happened, and recovery to a known-good state. Each of them requires an element that cannot be changed from software.
In Lenovo ThinkSystem servers, this role is played by a Root of Trust module with Platform Firmware Resiliency, described as part of Lenovo ThinkShield. According to the Lenovo Press guide for the SR650 V4, the module covers three images: UEFI, the XCC controller firmware and the code of the FPGA that controls the motherboard. For each of them it performs detection, i.e. verifying the authenticity of firmware and updates, and recovery of a corrupted image to a known-safe state.
The third function is protection: making sure that good firmware is not maliciously overwritten. Lenovo states that the functions are carried out by a separate security processor and that the module makes the server compliant with NIST SP 800-193. According to the manufacturer, the SR650 V4 is also compliant with NIST SP 800-147B, the BIOS protection guidelines for servers.
A compliance claim is a starting point, not the end of verification. It is worth asking which components the protection covers, because network adapters, RAID controllers and drives have their own firmware.
How do Secure Boot, TPM 2.0 and firmware verification differ?
They protect three different layers, and none replaces the others. Secure Boot is a UEFI feature that allows only signed operating system loaders to run. Lenovo notes that Secure Boot helps prevent malicious code from loading and attacks such as rootkit installation, and in V4 servers it can be enabled at the factory.
Secure Boot, however, is part of the UEFI firmware. If the firmware itself is modified, it stops being a reliable gatekeeper. This is where the root of trust comes in, checking the firmware before the firmware checks anything else.
TPM 2.0 is a hardware cryptographic module. It stores keys, for example for disk encryption, and records measurements of successive boot stages. Lenovo points out that the integrated TPM 2.0 enables digital signatures and remote attestation. This allows an external system to check whether the server booted in the expected configuration.
Together they form a chain: the root of trust verifies the firmware, the firmware with Secure Boot verifies the loader, and the TPM records what ran and protects the keys. Missing any link leaves a gap in a place the others do not cover.
What does this look like in practice on Lenovo ThinkSystem servers?
As a set of features, some of which are standard and some of which require an extended controller licence. In the latest Lenovo ThinkSystem V4 generation, for example the SR650 V4 (2U, Intel Xeon 6 processors), the standard includes XClarity Controller 3, a Root of Trust module with PFR, TPM 2.0 and Secure Boot. A chassis intrusion switch is also available.
The XCC3 Premier licence adds security-relevant features. System Guard monitors the hardware inventory for unexpected component changes and can log the event or prevent booting. For configure-to-order purchases, Lenovo can enable it at the factory and record the first configuration snapshot. Premier also brings modes that comply with CNSA and with FIPS 140-3, plus extended support for NIST SP 800-193.
Lenovo also describes supply chain safeguards. Suppliers are vetted through its Trusted Supplier Program, and firmware is digitally signed on hardware security modules (HSMs). In addition, the Intel Transparent Supply Chain service is available, documenting where components come from.
One note for mixed fleets. In V3-generation servers with AMD processors, such as the SR665 V3, the controller is the previous version, XCC2, with a different set of licences. Procedures, permissions and update schedules therefore have to be managed per generation, not for “Lenovo servers” as a whole.
Where should the management port sit, and who should have access to it?
On a separate management network, reachable only from designated administrative workstations. Lenovo V4 servers have a dedicated Ethernet port for controller management, and optionally a second one, installed in the OCP adapter slot. This port should go to a separate network or VLAN with no routing from user networks and no access from the internet.
The Lenovo Press guide for the SR650 V4 also describes a second option: the XCC controller can share one port on every OCP adapter installed in the server. That is convenient, because it saves cabling. It does mean, however, that controller traffic runs over the same physical path as production traffic. Such a configuration calls for a deliberate decision and careful logical separation, not accidental enablement.
Access to the management network is best routed through a jump server with multi-factor authentication. Controller accounts should be personal, and unused management protocols disabled. Controller logs have to be sent to a central system, because a BMC login at three in the morning is a security event, not an administrative one.
How do you update firmware across a server fleet without losing control over changes?
By treating firmware as a separate patch stream, with its own inventory and schedule. A server contains several layers of firmware: the BMC controller, UEFI, network adapters, storage controllers and drives. Each has its own versions and its own security advisories, and operating system patches cover none of them.
The first step is an inventory of versions across the whole fleet. The second is tracking the manufacturer’s security advisories and the CISA KEV catalogue for the models you own. The third is testing updates on a non-production server before rollout, because a firmware update is not always easy to reverse.
Lenovo provides firmware update tools and the XClarity One platform for central fleet management. The root of trust adds an important property here: an update is verified before it is written, so a tampered package should not reach the memory. We describe the process more broadly in the article on patch management, and our other materials in the vulnerability management knowledge base.
Does firmware security relate to NIS2 and the supply chain?
Yes, in two places in the Directive. Article 21(2)(d) of NIS2 requires supply chain security, including relationships with direct suppliers. The server manufacturer and its subcontractors supply firmware that runs with the highest privileges in your infrastructure. Its origin and the way it is signed are therefore part of the supplier assessment.
Article 21(2)(e) requires security in the acquisition, development and maintenance of network and information systems, including vulnerability handling. Buying a server is an acquisition, and updating firmware is maintenance. Questions about the root of trust, firmware signing and the BMC patch cycle therefore belong in the request for quote.
We describe supplier assessment under Polish rules in the article on high-risk suppliers under the Polish cybersecurity act, and explain the concept of the IT supply chain in our glossary.
How do you assess firmware security in a server offer?
By turning each requirement into a question for the vendor and a way to verify it. The table below orders these questions from the lowest layer to the process level.
| Requirement | Question for the vendor | How to verify |
|---|---|---|
| Hardware root of trust | Which firmware images does it protect, and can it restore a good image? | product guide, description of the components covered |
| NIST SP 800-193 compliance | Does the claim cover all three functions: protection, detection, recovery? | manufacturer documentation, required controller licence |
| Firmware signing | Where and with what are updates signed? | description of the signing process, e.g. on HSMs |
| Secure Boot and TPM 2.0 | Are they available, and can they be enabled at the factory? | configuration on delivery of the server |
| Hardware change control | Does the controller detect component changes and can it block booting? | test: swap a component on a test server |
| Management separation | Is there a dedicated BMC port, and how does the shared port work? | management network design before deployment |
| Protocols and accounts | Which protocols can be disabled, and can accounts be personal? | review of the controller configuration |
| Patch cycle | Where are BMC and UEFI security advisories published? | advisory subscription, comparison with CISA KEV |
| Supply chain | How does the manufacturer vet component suppliers? | description of the supplier vetting programme |
The order matters. The first rows decide whether the firmware can be trusted. The last ones decide whether that trust survives several years of operation.
Where should you start in an existing server room?
By finding every BMC controller and checking where it can be reached from. In many server rooms, some management interfaces ended up on user networks during installation and nobody moved them. Scanning the network for management interfaces and comparing the result with the server list usually brings surprises.
The second step is a review of accounts and versions: default passwords, shared accounts, BMC and UEFI firmware versions checked against the manufacturer’s advisories. The third is enabling Secure Boot where the operating system and drivers allow it, as well as chassis intrusion logging. The fourth is sending controller logs to the monitoring system.
The fifth step is testing. We verify server and management network configuration as part of our review of IT systems and infrastructure configuration, and the resilience of the firmware and BMC interfaces themselves in firmware penetration tests. For new purchases, we select configurations with security features as part of our server infrastructure deployments. We cover the basics of building a server room in the article on rack servers.
Checklist: 7 questions for your server vendor: which firmware the root of trust protects; whether it restores a good image; how updates are signed; whether Secure Boot and TPM can be enabled at the factory; whether the controller detects component changes; how to separate the BMC port; where security advisories are published.
Sources cited in the text: NIST SP 800-193 “Platform Firmware Resiliency Guidelines” (May 2018); NIST SP 800-147B “BIOS Protection Guidelines for Servers”; NVD, CVE-2024-54085, and the CISA Known Exploited Vulnerabilities catalogue (entry of 25 June 2025); Lenovo Press product guides for ThinkSystem SR650 V4, SR630 V4, SR665 V3 and SR850 V4 and the Lenovo Servers and Storage Portfolio Guide (as of September 2026); Directive (EU) 2022/2555 (NIS2), Article 21(2)(d) and (e).
Frequently asked questions
What is a BMC in a server?
A baseboard management controller is a separate processor on the server's motherboard, with its own software, network interface and accounts. It lets you power the server on and off remotely, mount media images, take over the console and update firmware — even when the operating system is not running. In Lenovo ThinkSystem V4 servers, this role is played by XClarity Controller 3.
Why is a firmware attack more dangerous than an operating system infection?
Because it works below the operating system. A change in UEFI or the controller firmware survives an OS reinstall and is out of reach of tools running inside the system, such as EDR. In SP 800-193 (May 2018), NIST notes that a successful attack on platform firmware can render the system inoperable, perhaps permanently, or require it to be reprogrammed by the manufacturer.
What is a hardware root of trust?
It is an independent hardware element that verifies the authenticity of firmware before it runs and during updates, protects it from unauthorised writes and can restore a known-good image. NIST describes these three functions — protection, detection and recovery — in the SP 800-193 Platform Firmware Resiliency guidelines.
Is Secure Boot enough to protect firmware?
No. Secure Boot makes sure only a signed operating system loader runs, but it is itself part of the UEFI firmware. If the firmware is modified, Secure Boot is no longer a trustworthy checkpoint. Protection of the firmware itself comes from a hardware root of trust, while TPM 2.0 stores keys and measurements.
Can the server management port sit on the production network?
It can, but it should not. The BMC interface gives full control over the server, so it should run on a separate management network, reachable from designated workstations with strong authentication. Some servers, such as the Lenovo SR650 V4, allow a network adapter port to be shared with the controller — a convenience that calls for a deliberate decision.
