Engineer working on a laptop connected to a device test rig, surrounded by wiring harnesses and test hardware

Firmware vulnerabilities & fixes

BMC, IPMI and Redfish vulnerabilities in embedded firmware: identify the stack, rank the CVEs, lock down the management plane

Last updated:

The BMC, IPMI and Redfish vulnerabilities that matter most as of 2026-11-17 start with CVE-2024-54085, a Redfish Host Interface authentication bypass in AMI MegaRAC SPx that CISA added to its KEV catalog on 2025-06-25, the first BMC flaw listed there. Its predecessors CVE-2023-34329 and CVE-2023-34330 follow, then two issues no version bump closes: the IPMI 2.0 RAKP hash leak (CVE-2013-4786) and the ASPEED pantsdown bridges (CVE-2019-6260). Identify the stack, update it, and switch off what the product does not need.

Key takeaways

  • A baseboard management controller is a second computer on the board: its own SoC, flash, RAM, Linux-based firmware and network port, powered whenever the chassis has standby power, even when the host is switched off.
  • Five CVEs cover the main patterns: a supplier code base with a KEV-listed bypass, two earlier flaws in the same code, a protocol-level weakness and a hardware design issue.
  • CVE-2024-54085 lets a network attacker pretend to be the host.
  • They are the two flaws that Eclypsium reported as BMC&C Part 2 and that AMI fixed in AMI-SA-2023006 on 18 July 2023.

What is a BMC, and why is it a separate attack surface?

A baseboard management controller is a second computer on the board: its own SoC, flash, RAM, Linux-based firmware and network port, powered whenever the chassis has standby power, even when the host is switched off. It lets an operator power-cycle the host, watch sensors, open a remote console and reflash the host BIOS or UEFI, so it sits below the operating system with more control than root on the host. Servers carry one, and so do many industrial PCs, storage appliances, network appliances and edge AI boxes built on server boards. The firmware usually comes from a supplier: AMI's MegaRAC SPx, the open-source OpenBMC project, or an OEM stack such as HPE iLO, Dell iDRAC or Supermicro's BMC firmware.

Eclypsium's write-up on the KEV listing sums up why this matters: BMC firmware "executes outside the scope of operating system (OS) controls, and has access to all resources" of the platform, so endpoint tools on the host do not see a compromised BMC. For a device maker the point is supply chain: one flaw in a BMC code base ships under many brands and model numbers.

BMCs expose three management interfaces, each with its own history of flaws:

  • IPMI over LAN (RMCP+ on UDP port 623): the older binary protocol, still enabled on many boards for tools such as ipmitool.
  • Redfish: DMTF's REST API over HTTPS, the modern replacement, served by the same web stack as the BMC's web interface.
  • Host-side interfaces: the paths from the host to the BMC, such as KCS or a USB virtual network for the Redfish Host Interface, and on some SoCs hardware bridges on LPC and PCIe.

This guide covers the BMC as a firmware component. For the host boot chain the BMC can reflash, see secure boot and hardware root of trust; for the generic bug class behind most BMC web flaws, see how to fix authentication bypass in embedded firmware.

Which BMC CVEs matter most for device makers?

Five CVEs cover the main patterns: a supplier code base with a KEV-listed bypass, two earlier flaws in the same code, a protocol-level weakness and a hardware design issue. Scores are NVD's primary CVSS and, for AMI's CVEs, AMI's own; EPSS values are FIRST's for 2026-11-16, and KEV status is from CISA's catalog version 2026.11.13.

CVE What it is Affected Fixed NVD / vendor CVSS EPSS KEV
CVE-2024-54085 Authentication bypass through the Redfish Host Interface by spoofing headers MegaRAC SPx 12.0 up to 12.7, 13.0 up to 13.5 SPx 12.7 and 13.5 (AMI-SA-2025003, 2025-03-11) 9.8 / 10.0 (v4.0) 0.607 Yes, added 2025-06-25
CVE-2023-34329 Authentication bypass by spoofing an HTTP header MegaRAC SPx 12 and 13 SPx 12.4 and 13.0 (AMI-SA-2023006, 2023-07-18) 8.0 / 9.1 0.012 No
CVE-2023-34330 Code injection through the Dynamic Redfish Extension interface MegaRAC SPx 12 and 13 SPx 12.4 and 13.2 (AMI-SA-2023006) 8.8 / 8.2 0.006 No
CVE-2013-4786 IPMI 2.0 RAKP message 2 hands out a password HMAC for offline cracking Any BMC with IPMI 2.0 over LAN No protocol fix; mitigate in configuration 7.5 0.984 No
CVE-2019-6260 (pantsdown) ASPEED AST2400/AST2500 AHB bridges give the host read and write access to BMC memory Aspeed-based BMCs, incl. OpenBMC up to at least 2.6, AMI and Supermicro stacks Firmware and platform configuration that disables the bridges 9.8 0.042 No

The EPSS column ranks differently from the KEV column. CVE-2013-4786 scores 0.984 because it needs nothing but a reachable IPMI port and a user name, while CVE-2024-54085 at 0.607 is the one CISA confirms exploited. Our KEV and EPSS prioritisation guide explains how to weigh both signals, and our advisory on CVE-2024-54085 lists the affected NetApp products.

How does the MegaRAC Redfish bypass (CVE-2024-54085) work?

CVE-2024-54085 lets a network attacker pretend to be the host. The DMTF Redfish Host Interface specification (DSP0270 1.3.1) defines a TCP/IP path, for example a USB network connection, from the host's operating system to the BMC's Redfish service, and lets implementations offer an optional, configurable "AuthNone" mode for that path. The BMC therefore has to decide whether a request arrived from the host, and MegaRAC got that decision wrong.

Eclypsium's Part 3 analysis traces the bug to a Lua module under /usr/local/redfish/extensions/host-interface/ that compares the X-Server-Addr header, or Host if absent, against the host interface address stored in the BMC's database, 169.254.0.17 by default. Only the text up to the first colon is compared, so a remote request carrying X-Server-Addr: 169.254.0.17: passes as a host request; on an HPE Cray XD670 running firmware 1.17 Eclypsium created a new administrator account through Redfish this way. It also reported that the flaw appeared after the CVE-2023-34329 fix, with the same impact, and confirmed it on HPE Cray XD670 1.09, 1.13 and 1.17, an Asus RS720A-E11-RS24U at 1.2.27 and an ASRock Rack device, and found roughly 1,000 instances exposed on the internet through Shodan.

AMI-SA-2025003, first published on 11 March 2025, fixes it in SPx_12.7 and later and in SPx_13.5. AMI ships those fixes to the OEMs that license MegaRAC, and each OEM releases its own firmware, so Eclypsium's patch list names AMI, Lenovo and HPE advisories separately. If your product uses a MegaRAC-based board, the fix you need is the board vendor's build with SPx 12.7 or 13.5 code underneath, and you should get that mapping in writing.

What were CVE-2023-34329 and CVE-2023-34330?

They are the two flaws that Eclypsium reported as BMC&C Part 2 and that AMI fixed in AMI-SA-2023006 on 18 July 2023. CVE-2023-34329 is the earlier header-spoofing authentication bypass, fixed in SPx_12.4 and SPx_13.0; AMI classes it CWE-306 and NVD CWE-290. CVE-2023-34330 is code injection through the Dynamic Redfish Extension interface, fixed in SPx_12.4 and SPx_13.2. The same advisory carries five CVEs from Nozomi Labs, fixed between SPx_12.2 and 12.5 and SPx_13.0 and 13.3, including a hard-coded credential (CVE-2023-34473, CWE-798).

The two scorers disagree on both. AMI rates CVE-2023-34330 at 8.2 with local access and high privileges, NVD at 8.8 from the network with low privileges; for CVE-2023-34329 AMI gives 9.1 with network access and NVD 8.0 from an adjacent network. The practical reading for a device maker: on a SPx 12 build below 12.7 or a SPx 13 build below 13.5 you are exposed to the KEV-listed bypass whatever happened to the 2023 fixes, so move to the 2025 fix level directly.

Why can't the IPMI RAKP hash leak (CVE-2013-4786) be patched?

Because it is how the protocol works. NVD describes CVE-2013-4786 as a property of the IPMI 2.0 specification's RMCP+ Authenticated Key-Exchange Protocol: a remote attacker can obtain password hashes "by obtaining the HMAC from a RAKP message 2 response from a BMC". The original write-up on cracking IPMI passwords shows that the client can send a deliberately wrong password and still receive the HMAC, and that the response also reveals whether the account name exists. In a scan of a bit over 1,000 internet-facing BMCs, a bit over 5% still had vendor default passwords and another 5 to 10% of accounts fell to a modest dictionary. Its advice is to keep BMCs behind firewalls and block UDP port 623.

Cipher suite 0 is a separate IPMI 2.0 failure in the same family. It provides no authentication, integrity or confidentiality, and the cipher zero write-up shows an iDRAC answering ipmitool -I lanplus -C 0 commands, such as a user list, with an arbitrary password for a valid user name. A device maker cannot fix the protocol but controls the defaults, which the hardening section below covers.

What is pantsdown (CVE-2019-6260)?

Pantsdown is a set of hardware bridges in ASPEED's AST2400 and AST2500 BMC SoCs that let the host read and write the BMC's physical memory. Stewart Smith's disclosure lists eight paths, including the iLPC2AHB bridge, the PCIe VGA P2A bridge, X-DMA and a UART debug shell, several "enabled at cold start". The consequences listed include reflashing or dumping the running BMC firmware from the host and bricking the BMC until the next AC power cycle, which matters wherever host and BMC sit in different trust domains, such as a leased bare-metal server or an appliance whose host runs customer code.

The disclosure names OpenBMC, AMI's BMC firmware and Supermicro as affected stacks, OpenBMC up to at least 2.6 on all supported Aspeed platforms. The fix is platform-specific and can need both BMC and host firmware changes: OpenBMC added a U-Boot patch that disables the bridges, opt-in per platform, because some designs use the bridges as a feature. One trap is that disabling the iLPC2AHB bridge removes only write access; reads continue until SuperIO decoding is turned off, which in turn removes host access to SUARTs, GPIOs and the host-BMC mailbox. Our U-Boot bootloader security guide covers the bootloader layer where this configuration lives.

How do you identify the BMC stack and version in a firmware image?

Treat the BMC image as its own firmware with its own SBOM, separate from the host BIOS and operating system.

  1. Find the BMC imageIt usually arrives as a separate update file from the host BIOS. Unpack it as you would any embedded Linux image: U-Boot, a kernel and a SquashFS or similar root file system.
  2. Read the identity filesOpenBMC builds write /etc/os-release from an os-release recipe that adds OPENBMC_TARGET_MACHINE, EXTENDED_VERSION and BUILD_ID, with the version taken from git describe of the build tree, so you get the machine name and the upstream tag the vendor forked. MegaRAC SPx images carry AMI's Redfish service; the /usr/local/redfish tree Eclypsium describes is one marker.
  3. Map the OEM version to the supplier releaseAMI's advisories use SPx release numbers such as 12.7 and 13.5, but products ship OEM versions such as Cray XD670 1.17. Ask the board vendor which SPx code level each build contains; a scanner cannot derive it from the OEM number.
  4. Record the SoCAn AST2400 or AST2500 brings CVE-2019-6260 into scope whatever the firmware vendor, and the fix depends on how the platform configures the bridges.
  5. Read the enabled servicesCheck whether the IPMI network daemon is installed and started, which cipher suites its configuration permits, whether Redfish Host Interface is enabled and with which authentication modes, and which default accounts exist.

How should a device maker configure the BMC?

Most BMC exposure comes from interfaces left on for convenience. A device baseline:

  • Keep the BMC off untrusted networks. Put the dedicated management port on an isolated management network, never port-forward it, and document that in the product's installation guide. Our CISA BOD 23-02 guide explains why management interfaces reachable from the internet are treated as an incident.
  • Disable IPMI over LAN unless the product needs it. If it must stay on, permit only strong cipher suites. OpenBMC's default cipher list enables only cipher suite 17. On other stacks the ipmitool manual describes lan set <channel> cipher_privs, a 15-character string where each position is a cipher suite and X marks it unused, so the first character must be X to turn off cipher 0.
  • Ship no shared default password. Use a unique random password per unit or force a change at first login, and remove or rename fixed vendor accounts. Our hard-coded credentials guide covers per-device secrets.
  • Turn off the Redfish Host Interface if the host does not use it, and never enable AuthNone on it. Where the host needs it, use credentials as DSP0270 requires, valid on the external interface or auto-generated for the host.
  • Close the hardware paths. On ASPEED SoCs, apply the vendor's pantsdown mitigations for every bridge the platform does not use.
  • Enable signed BMC updates. The BMC can reflash the host, so a BMC that accepts unsigned images undoes the host's secure boot. NIST SP 800-193 treats the BMC as one of the platform firmware devices that needs protection, detection and recovery.

How do you patch and test BMC firmware?

  1. Get the fixed supplier codeFor MegaRAC, that means a board vendor build containing SPx 12.7+ or 13.5 for CVE-2024-54085; for OpenBMC, rebase on a current release or backport the upstream fixes.
  2. Verify the fix in the shipped buildConfirm the version mapping in the vendor's advisory, and for a backport compare the patched files. Our firmware diffing and patch analysis guide shows how.
  3. Test the interfaces on a unit, not only the imageSend a Redfish request with a spoofed X-Server-Addr header and confirm it is rejected; run ipmitool -I lanplus -C 0 with a wrong password and confirm the session fails; confirm UDP port 623 is closed when IPMI over LAN is off; on ASPEED platforms, check from the host that the bridges are closed.
  4. Recheck after factory reset and BMC rebootPantsdown's disclosure notes that a watchdog reset in "System Reset" mode reopens every bridge enabled at cold start until firmware reconfigures them, and settings can return to defaults after a reset.
  5. Publish per-CVE statusRegenerate the BMC SBOM, record the supplier release and OEM version, and state for each CVE whether the product is affected. Our CSAF and VEX for device makers guide shows the format.

Worked example: how would you triage the BMC in an industrial PC image?

Synthetic example, not a real product. An industrial PC maker scans the BMC update file for a box built on a third-party server board.

  1. Identify the stackThe root file system contains /usr/local/redfish with Lua modules and lighttpd, consistent with MegaRAC SPx. The OEM version is 2.04; the board vendor confirms in writing that it is built on SPx 13.3. The SoC is an AST2500.
  2. Match the CVEsCVE-2024-54085 applies (13.3 is below 13.5) and is in KEV. CVE-2023-34329 and CVE-2023-34330 are fixed (both fixed at or below 13.2). CVE-2013-4786 applies because IPMI over LAN is on. CVE-2019-6260 applies until the vendor confirms which ASPEED bridges the build disables.
  3. Check configurationIPMI cipher 0 is permitted at administrator level, the Redfish Host Interface is enabled, and the BMC ships with a fixed default password printed in the manual.
  4. FixAsk the board vendor for a SPx 13.5-based build, disable IPMI over LAN and cipher 0 by default, disable the Redfish Host Interface since the product's host software does not use it, introduce per-unit passwords, then rebuild, re-scan, retest on a unit and publish a VEX statement per CVE.

How does AI-assisted analysis help with BMC findings?

SoftWSP's automated scan unpacks a firmware image, catalogs the software packages it can identify in the extracted file system with their versions (the inventory our firmware SBOM guide describes) and matches each one against CVE records, CISA's KEV catalog and FIRST's EPSS scores. In a BMC image that covers the open-source layers, such as the kernel, U-Boot, the web server and OpenSSL. This is deterministic code: the code scores, and AI narrates. You can inspect the resulting CycloneDX file in the free SBOM viewer.

The proprietary layer is where version matching stops. NVD's configuration for CVE-2023-34329 and CVE-2023-34330 lists MegaRAC SPx only as versions 12 and 13, an OEM version number does not reveal the SPx code level beneath it, and pantsdown depends on the SoC and its bridge settings, not on a package version. AI's part is the written explanation of what matched, citing the component, version and record behind each sentence. An engineer reviews every finding before the report is delivered, and that review is where the supplier release, the SoC, the enabled IPMI cipher suites, the Redfish Host Interface setting and any vendor backport become a documented "affected" or "fixed" decision.

For where the BMC fits in a full assessment, see the OWASP Firmware Security Testing Methodology guide, or device & firmware security services to have a specific BMC image reviewed.

This guide draws on AMI security advisories AMI-SA-2023006 and AMI-SA-2025003, Eclypsium's BMC&C Part 3 analysis and KEV post, Stewart Smith's CVE-2019-6260 disclosure, the fish2.com IPMI RAKP and cipher zero write-ups, DMTF DSP0270 1.3.1, OpenBMC's IPMI cipher list and os-release recipe, the ipmitool manual, NVD, CISA's KEV catalog (version 2026.11.13) and FIRST EPSS data dated 2026-11-16, all read directly on 2026-11-17.

Put this guide to work

Frequently asked questions

Which BMC vulnerabilities matter most for device makers?

As of 2026-11-17, start with CVE-2024-54085, an authentication bypass in the Redfish Host Interface of AMI MegaRAC SPx that AMI fixed in SPx 12.7 and 13.5 and that CISA added to its KEV catalog on 2025-06-25. It is the only BMC entry in KEV catalog version 2026.11.13. Next come its predecessors CVE-2023-34329 (header-spoofing authentication bypass) and CVE-2023-34330 (code injection through the Dynamic Redfish Extension interface), then two older issues that version upgrades alone do not close: CVE-2013-4786, the IPMI 2.0 RAKP password-hash disclosure, and CVE-2019-6260, the ASPEED AST2400/AST2500 host-to-BMC bridges known as pantsdown.

How do I tell which BMC firmware stack is in an image?

Unpack the BMC flash image and read its root file system. OpenBMC builds write an os-release file with fields such as OPENBMC_TARGET_MACHINE and EXTENDED_VERSION, the git describe output of the build. AMI MegaRAC SPx images carry AMI's own Redfish service; Eclypsium's analysis of CVE-2024-54085 located the vulnerable code in a Lua module under /usr/local/redfish. Also record the BMC SoC, for example an ASPEED AST2400 or AST2500, and the OEM firmware version, which rarely matches the SPx release number in AMI's advisories.

Can the IPMI RAKP hash disclosure (CVE-2013-4786) be patched?

Not by a firmware update to the protocol itself. NVD describes it as a property of the IPMI 2.0 specification: a BMC answers RAKP message 1 with an HMAC of the user's password in RAKP message 2, which anyone can crack offline. Mitigation is configuration: turn off IPMI over LAN when the product does not need it, never expose UDP port 623 beyond a management network, use long random per-device passwords and remove or rename default accounts.

What is IPMI cipher suite 0 and how do I disable it?

Cipher suite 0 is the IPMI 2.0 suite with no authentication, integrity or confidentiality, so a BMC that permits it at administrator level accepts commands with any password for a valid user name. With ipmitool, the lan set cipher_privs command maps each of the 15 cipher suites to a maximum privilege, and X marks a suite as unused, so the first character must be X. OpenBMC's default IPMI configuration enables only cipher suite 17.

Does SoftWSP find BMC vulnerabilities in my firmware automatically?

Partly. SoftWSP's automated scan unpacks a firmware image, catalogs the software packages it can identify with their versions, such as the Linux kernel, U-Boot, lighttpd or OpenSSL inside a BMC image, and matches them against CVE records, CISA's KEV catalog and FIRST EPSS scores. AI then writes a plain-language summary citing the component, version and record behind each finding, and an engineer reviews it. Identifying the MegaRAC SPx or OpenBMC release, the OEM build it maps to, and whether IPMI over LAN, cipher 0 or the Redfish Host Interface are enabled is part of that engineer review.

Sources

  1. AMI Security Advisory AMI-SA-2025003 (CVE-2024-54085)
  2. AMI Security Advisory AMI-SA-2023006, MegaRAC SPx (CVE-2023-34329, CVE-2023-34330)
  3. Eclypsium: BMC&C Redfish Alert 3, AMI MegaRAC vulnerabilities part 3
  4. Eclypsium: BMC vulnerability CVE-2024-54085 added to CISA's KEV catalog
  5. Stewart Smith: CVE-2019-6260, gaining control of BMC from the host processor
  6. Cracking IPMI passwords remotely (RAKP hash disclosure)
  7. The infamous IPMI cipher zero
  8. DMTF DSP0270 Redfish Host Interface Specification 1.3.1
  9. OpenBMC: default IPMI cipher suite list
  10. OpenBMC: os-release recipe append
  11. ipmitool manual page (lan set cipher_privs)
  12. NIST NVD: CVE-2024-54085
  13. NIST NVD: CVE-2023-34329
  14. NIST NVD: CVE-2023-34330
  15. NIST NVD: CVE-2013-4786
  16. NIST NVD: CVE-2019-6260
  17. CISA: Known Exploited Vulnerabilities (KEV) Catalog
Was this guide helpful?

In this cluster

BMC, IPMI and Redfish vulnerabilities in embedded firmware: identify the stack, rank the CVEs, lock down the management plane