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

Firmware vulnerabilities & fixes

NIST SP 800-193 (Platform Firmware Resiliency Guidelines), explained

Last updated:

NIST SP 800-193, "Platform Firmware Resiliency Guidelines," published as Final in May 2018 and never revised since, is the primary free, vendor-neutral reference for building device and system firmware that can survive a destructive attack rather than just resist one. Authored by Andrew Regenscheid of NIST with technical input from engineers across Cisco, Dell, HPE, HP Inc., IBM, Intel, Microsoft and the NSA. Its own abstract frames the stakes directly: "A successful attack on platform firmware could render a system inoperable, perhaps permanently, or requiring reprogramming by the original manufacturer, resulting in significant disruptions to users." The guidelines are written for "system and platform device vendors," but explicitly extend "to other classes of systems with updatable firmware, including Internet of Things devices, embedded systems, and mobile devices" — scope that reaches well beyond the PC/server/network-device platforms used in its own worked examples.

Key takeaways

  • Most firmware security work focuses on finding a specific exploitable bug — a command injection or hard-coded credential in a running image.
  • Recoverable — every critical device meets the detection guidelines (Sections 4.1/4.3) and the recovery guidelines (Sections 4.1/4.4).
  • User-initiated updates — vendor-supplied utilities run from external media or the host OS.
  • Section 4 uses RFC 2119 wording, so "shall" is the mandatory floor and "should" a recommendation.

Why should a firmware security assessment care?

Most firmware security work focuses on finding a specific exploitable bug — a command injection or hard-coded credential in a running image. SP 800-193 asks a different, complementary question: when an attack on firmware succeeds anyway, can the device detect the corruption and recover without being bricked or physically reprogrammed by the manufacturer? That question matters for exactly the kind of multi-component embedded systems a firmware SBOM inventories — a platform's BIOS, embedded controller, BMC/management engine, NIC and GPU firmware can each have their own update path, and SP 800-193's guidance is organized device-by-device for precisely that reason.

What are the three principles: Protection, Detection, Recovery?

Section 3.1 states the security guidelines "are based on the following three principles," and that the technical guidelines in Section 4 "are organized around these principles":

Principle What it covers
Protection Ensuring firmware code and critical data "remain in a state of integrity and are protected from corruption," primarily through authenticity and integrity checks on firmware updates
Detection Mechanisms for detecting when firmware code and critical data "have been corrupted or otherwise changed from an authorized state"
Recovery Mechanisms for "restoring Platform Firmware code and critical data to a state of integrity" once corruption is detected, or when an administrator forces recovery

The document is explicit that Protection alone is not assumed to be sufficient: "While protection mechanisms are intended to prevent destructive or malicious attacks against platform firmware and critical data, these mechanisms may be imperfect or impractical to implement on all categories of devices. In those cases, detection and recovery mechanisms are intended to discover and remediate attacks to regain normal and secure operation on the device." Section 3.1 also notes Protection's scope "is similar in... purpose to the guidelines found in NIST SP 800-147, BIOS Protection Guidelines" — SP 800-193 generalizes that earlier, BIOS-only document's approach to the full set of platform devices.

What are the three resiliency tiers: Protected, Recoverable, Resilient?

Section 3.2 defines platform-wide claims built from device-level guidelines, anchored to a platform's critical platform devices — those "necessary to minimally restore operation of the system, and sufficient to restore reasonable functionality":

  • Protected — every critical device meets the Section 4.1/4.2 protection guidelines, but "may not fully offer capabilities to recover the device's firmware and/or critical data."
  • Recoverable — every critical device meets the detection guidelines (Sections 4.1/4.3) and the recovery guidelines (Sections 4.1/4.4).
  • Resilient — every critical device meets all of Section 4 (protection, detection and recovery together), and non-critical devices either meet the same bar or are isolated so that "a compromise of one of these devices will not impact the security of the platform as a whole."

A platform can be Protected without being Recoverable, and the tiers are not a strict ladder on every axis — Protected has no recovery obligation at all, which makes it a meaningfully weaker (and cheaper to implement) claim than Recoverable or Resilient.

What are the Roots of Trust and firmware update mechanisms?

Section 4.1 organizes the rest of the guidelines around three logical components, each possibly distinct or shared: the Root of Trust for Update (RTU), responsible for authenticating firmware updates; the Root of Trust for Detection (RTD), responsible for corruption-detection capability; and the Root of Trust for Recovery (RTRec), responsible for recovery once corruption is detected or an administrator forces it. Section 3.3 defines a Root of Trust as an element that "must be designed to always behave in the expected manner because its proper functioning is essential to providing its security-specific functions and because its misbehavior cannot be detected."

Section 3.5 separates authentication (verifying an update image's source and integrity via digital signature, rooted in the RTU's key store) from authorization (permission to perform the update, "typically rooted in the device or system owner" rather than the manufacturer). Section 3.5.2 names five authorized update mechanisms:

  1. User-initiated updates— vendor-supplied utilities run from external media or the host OS.
  2. Managed updates— remote administrator-driven updates via management agents.
  3. Rollback— reverting to an earlier signed firmware version, explicitly framed as a defence: "this mechanism guards against attackers installing old firmware with known vulnerabilities" (implying rollback itself needs its own safeguards against abuse).
  4. Manual recovery— a physically-present user replacing a corrupted image with a known-good one.
  5. Automatic recovery— the device detecting corruption and restoring from a separately stored backup image on its own.

Section 3.5.3 adds secure local update, which authorizes a firmware change "through a process demonstrating unambiguous physical presence" instead of a signature chain — useful for recovering an image that cannot be reached through the authenticated-update path, but flagged with a sharp caveat: "remote mechanisms, such as interacting with a device or system via a remote console, do not satisfy this requirement for physical presence," and devices relying on it are themselves "potentially vulnerable to attacks by rogue administrators or other attacks with physical access to the device."

What do the Section 4 "shall" requirements say, device by device?

Section 4 uses RFC 2119 wording, so "shall" is the mandatory floor and "should" a recommendation. The table below lists the requirements a device maker most often has to design for, from the Final PDF:

Section Requirement (paraphrased, "shall" unless noted)
4.1.1 All RoTs and CoTs are immutable or integrity-protected; information from host software to platform firmware is treated as untrusted
4.1.2 Each device with mutable firmware relies on an RTU or a chain anchored in one; the RTU holds a key store and a FIPS 186-4 signature check; an RTU update survives power loss mid-write
4.2.1.1 Update images are signed with at least 112-bit security strength and verified before the non-volatile write completes
4.2.1.2 Flash regions holding firmware are writable only through the authenticated update or a secure local update
4.2.1.3 The update path cannot be bypassed, and can block unauthorized rollback to a version with a known weakness
4.2.4 Critical data changes only through defined interfaces, is validated before commit, and factory defaults are protected "at least as well as" code
4.3.1 The RTD verifies code by signature or hash before code outside it runs; FPGAs and CPLDs without crypto use hardware load-failure detection
4.4.1 Recovery resists the same attack that corrupted the primary image; the recovery image is protected and verified by the authenticated update mechanism
4.4.2 The RTRec can restore critical data to factory defaults, and should be able to restore the last known good copy

Logging, notification and administrator-forced recovery are mostly "should" items (4.3.1 items 4–5, 4.4.1 items 8–12), so two Recoverable products can differ a lot in what the operator sees.

How do symbiont and host devices share a Root of Trust?

Many components in a device — a Wi-Fi module, a sensor hub, a storage controller — cannot verify or recover their own firmware. Section 3.4 calls them symbiont devices and lets a host device lend the missing capability: "a host device and a symbiont device jointly fulfill the guidelines for protection, detection and/or recovery that the symbiont device cannot fulfill independently." The host carries extra weight: Section 4 says it "must additionally meet all of the requirements for a self-contained device," and Section 4.1.1 requires a secure channel when a chain of trust crosses device boundaries.

How should recovery be automated without an attack/recovery cycle?

Section 3.6.4 names three recovery modes — fully automated, partially automated, and manual — and flags a specific failure pattern worth calling out to anyone designing a recovery policy: "an administrator-defined policy may constrain the versions of firmware that may be installed during the recovery. Those who set recovery policies must do so with care. The firmware version set in the policy might well be the version that was successfully attacked necessitating the recovery. Simply rewriting the vulnerable version may lead to an attack/recovery cycle." The same section also recommends that recovery schemes "provide methods to retain or record attacked images and other information" where practical, since overwriting firmware in place can destroy forensic evidence of how the attack happened.

How does SP 800-193 relate to SP 800-147, SP 800-155 and TCG specifications?

SP 800-193 is the newest of four NIST firmware documents, and it leans on Trusted Computing Group (TCG) work for the hardware side. The mapping below is SoftWSP's reading; SP 800-193's own reference list cites SP 800-147, SP 800-147B and the TCG TPM Library, but not SP 800-155.

Document Status Covers Maps to
SP 800-147, BIOS Protection Guidelines Final, April 2011 Authenticated BIOS updates on client PCs (see our SP 800-147 guide) Protection, host boot firmware only
SP 800-147B, BIOS Protection Guidelines for Servers Final, August 2014 The same model for server BIOS and service processors Protection, servers
SP 800-155, BIOS Integrity Measurement Guidelines Draft, 8 December 2011 (the only version on CSRC) A BIOS "integrity measurement and reporting chain" for enterprise desktops and laptops Detection, plus remote reporting that SP 800-193 leaves out
TCG TPM 2.0 Library Version 185, March 2026; published as ISO/IEC 11889:2015 Key storage, measurements and attestation; SP 800-193 lists the TPM as a platform device whose firmware must be protected Measurement and attestation evidence for Detection
TCG Cyber Resilient Module and Building Block Requirements V1 R0.2, August 2022 Building blocks "such as watchdogs and latches" and module profiles Protect, detect and recover on a single module
TCG DICE Specification family A cryptographic identity for constrained devices with minimal hardware A small RoT for devices without a TPM

In short, SP 800-193 sets the properties and TCG describes hardware that can provide them. For the boot chain, see secure boot and hardware root of trust.

Worked example: how would you grade a connected gateway?

Synthetic example, not a real product. A cellular industrial gateway has an application SoC booting from SPI NOR flash, a cellular modem with its own firmware, a Wi-Fi module, and an eMMC holding configuration.

  1. List the critical devices (Section 3.2)The SoC boot firmware, the SPI flash and the eMMC configuration are needed to restore operation. The modem is needed for the product's purpose. The Wi-Fi module is not, if the gateway works without it.
  2. Check Protection (4.1, 4.2)The SoC ROM verifies a signed bootloader, and the bootloader verifies the OS image: the SoC has an RTU. The modem accepts only vendor-signed images: it has its own RTU. The configuration on eMMC is written by the web UI with no validation: 4.2.4 fails.
  3. Check Detection (4.3)The SoC detects a bad image at boot; the modem only reports a version string to the SoC, which is not integrity verification.
  4. Check Recovery (4.4)An A/B partition scheme restores the OS, but both copies sit in the same writable flash region with no independent protection, so 4.4.1 item 1 is in doubt. No factory-default restore exists for configuration (4.4.2 item 5).
  5. Grade itThe gateway does not reach Protected, because one critical device (the configuration store) fails 4.2.4. Validating configuration writes and protecting factory defaults gets it to Protected. Adding modem detection through the SoC as host, and a write-protected recovery image, is the route to Recoverable.

What are the common pitfalls?

  • Treating critical data as "just config". Section 4.2.4 puts configuration, MAC addresses and factory defaults in scope; unsigned or unvalidated settings can stop a device from booting.
  • Debug and diagnostic paths. Section 4.2.1.3 lists "development or diagnostic interfaces that allow access to flash regions" as bypasses to disable in production; see JTAG and SWD debug port security.
  • Unbounded rollback. A signed old image still has its old bugs; 4.2.1.3 asks the update mechanism to block unauthorized rollback to a version with a known weakness.
  • Recovery image beside the primary. If one flash write-enable exposes both copies, the attack that kills the primary kills the backup.
  • Power loss mid-update. Section 4.1.2 requires the RTU to stay operational or recoverable after "power loss in the middle of a flash write operation."
  • Key store with no recovery plan. A non-updatable key store needs a plan for a leaked signing key, as the note under 4.1.2 says.

Where should a device maker start?

  1. Inventory which devices in your platform are "critical"under Section 3.2's definition before claiming any tier — a platform is only as resilient as its weakest device that's actually needed to boot or restore function.
  2. Treat Protection, Detection and Recovery as three separate gaps to close, not one. A device with excellent signature-verified updates (Protection) but no way to detect silent corruption or recover from it (no Detection/Recovery) can still only claim "Protected," not "Recoverable" or "Resilient."
  3. Audit rollback and recovery policies for the attack/recovery-cycle trapdescribed in Section 3.6.4 — a recovery mechanism that silently restores a known-vulnerable firmware version is not a fix.
  4. Use SP 800-193's device list (Section 2.1) as a checklistof components that often get missed in a firmware inventory beyond the main boot firmware: embedded controller/Super I/O, TPM, BMC/management engine, NIC and GPU expansion ROMs, and SPI flash.
  5. Write the update path down onceand check it against firmware OTA update security practice, so signing, rollback and recovery rules match across every device.

Root-of-trust design, signature checks and recovery behaviour need an engineer's review against the hardware. SoftWSP covers the software inside the image: the platform extracts a firmware SBOM from the binary as CycloneDX 1.6 JSON, matches the components it identifies against NVD, CISA KEV, FIRST EPSS and CISA ICS advisories, and drafts a cited report that an engineer reviews. That gives you the known-vulnerability picture for the bootloader, OS and libraries you are about to make recoverable; it does not test whether the device meets SP 800-193. See Device & Firmware Security.

This guide summarises NIST SP 800-193 (May 2018, Final); it is not a substitute for the full text and does not constitute compliance or legal advice.

Put this guide to work

Frequently asked questions

What is NIST SP 800-193?

"Platform Firmware Resiliency Guidelines," a NIST Special Publication authored by Andrew Regenscheid of NIST, with acknowledged technical contributions from engineers at Cisco, Dell, Hewlett Packard Enterprise, HP Inc., IBM, Intel, Microsoft and the National Security Agency. Its own abstract states it "provides technical guidelines and recommendations supporting resiliency of platform firmware and data against potentially destructive attacks." It was published as Final in May 2018 (finalized 4 May 2018, after a draft dated 30 May 2017), carries no later revision as of this guide's writing, and is available free of charge at its DOI, https://doi.org/10.6028/NIST.SP.800-193. Its stated audience is "system and platform device vendors of computer systems, including manufacturers of client, servers, and networking devices," while noting its principles "should be broadly applicable to other classes of systems with updatable firmware, including Internet of Things devices, embedded systems, and mobile devices."

What are the three resiliency principles in SP 800-193?

Section 3.1 states the guidelines "are based on the following three principles": Protection — "Mechanisms for ensuring that Platform Firmware code and critical data remain in a state of integrity and are protected from corruption, such as the process for ensuring the authenticity and integrity of firmware updates"; Detection — "Mechanisms for detecting when Platform Firmware code and critical data have been corrupted or otherwise changed from an authorized state"; and Recovery — "Mechanisms for restoring Platform Firmware code and critical data to a state of integrity in the event that any such firmware code or critical data are detected to have been corrupted, or when forced to recover through an authorized mechanism." The document notes protection mechanisms "may be imperfect or impractical to implement on all categories of devices," which is exactly why detection and recovery exist as separate, independently gradable properties rather than protection alone being the target.

What do 'Protected', 'Recoverable' and 'Resilient' mean for a platform?

Section 3.2 defines three platform-level resiliency properties built on top of a document-wide concept of 'critical platform devices' (the devices "necessary to minimally restore operation of the system, and sufficient to restore reasonable functionality"). Protected: "all critical platform devices must meet the protection guidelines found in Sections 4.1 and 4.2, but may not fully offer capabilities to recover the device's firmware and/or critical data." Recoverable: "all critical platform devices must provide the means to detect corruption... and provide the means to recover from this corruption." Resilient: "all critical platform devices must meet all of the guidelines in Section 4," with non-critical devices expected to meet the same bar or be isolated so a compromise of one "will not impact the security of the platform as a whole." The three are not synonyms for 'better' in sequence on every axis — Protected has no recovery requirement at all, while Recoverable and Resilient both require it.

What firmware update mechanisms does SP 800-193 describe?

Section 3.5 splits updates into authentication (verifying a firmware image's source and integrity via digital signatures, rooted in a Root of Trust for Update holding a public key or its hash) and authorization (permission to perform the update, typically rooted in the device or system owner, not the manufacturer). Section 3.5.2 lists five authorized update mechanisms: user-initiated updates, managed (remote administrator) updates, rollback to an earlier signed version as a defence against reintroducing known vulnerabilities, manual recovery requiring physical presence, and automatic recovery from a separately stored backup image. Section 3.5.3 adds secure local update — authorizing an update "through a process demonstrating unambiguous physical presence" rather than a signature chain — with an explicit warning that "remote mechanisms... do not satisfy this requirement for physical presence" and that such local-update paths are themselves "potentially vulnerable to attacks by rogue administrators or other attacks with physical access to the device."

Sources

  1. NIST: SP 800-193, Platform Firmware Resiliency Guidelines — publication page
  2. NIST SP 800-193 (Andrew Regenscheid), full text PDF
  3. NIST: SP 800-147, BIOS Protection Guidelines (the earlier, narrower-scope document SP 800-193 builds on)
  4. NIST: SP 800-147B, BIOS Protection Guidelines for Servers (Final, August 2014)
  5. NIST: SP 800-155 (Draft), BIOS Integrity Measurement Guidelines — publication page and document history
  6. Trusted Computing Group: TPM 2.0 Library Specification
  7. Trusted Computing Group: Cyber Resilient Module and Building Block Requirements
  8. Trusted Computing Group: Device Identifier Composition Engine (DICE) work group
Was this guide helpful?

In this cluster