[Report] Shade Core: The Resurgence of UEFI Attacks on Modern PCs Protected by Hypervisors - CODE BLUE 2026 #codeblue_jp #codeblue2026

[Report] Shade Core: The Resurgence of UEFI Attacks on Modern PCs Protected by Hypervisors - CODE BLUE 2026 #codeblue_jp #codeblue2026

A report on the CODE BLUE 2026 session "Shade Core: The Revival of UEFI Attacks on Modern PCs Protected by Hypervisors."
2026.11.18

This page has been translated by machine translation. View original

Hello, this is Usuda.

Everyone, are you taking security measures? (greeting)

This time, this is a report on the following session held at CODE BLUE 2026.

The session introduced techniques for attack methods that do not depend on hypervisors.

Shade Core: Revival of UEFI Attacks on Modern PCs Protected by Hypervisors

UEFI malware is a serious threat to enterprise endpoints in contexts such as national security. However, with advances in platform security, many existing UEFI attacks have lost their effectiveness against modern PCs. Hypervisor-based security (HBS) is now enabled by default, and the privileges of SMM (System Management Mode) have also been stripped (reduced), so there are almost no attacks that can bypass HBS. Even the few such attacks are unstable, depend on specific hypervisors, and can easily be invalidated by implementation changes.

In this talk, we propose "Shade Core," a general-purpose technique for bypassing HBS under any hypervisor. We also introduce "Shading Attacks," a new class of BIOS attacks that encompasses Shade Core. Unlike conventional BIOS attacks that tamper with components such as the OS or hypervisor, Shading Attacks have the distinctive feature of not requiring modification of any post-BIOS component.

This presentation provides analysts with a systematic understanding of the attack capabilities of BIOS malware against modern PCs, and offers defenders deep insights (insight) for detecting this new class of attacks.

Speakers: Kazuki Matsuo 松尾 和輝

Report

  • A discussion on how to defend when system firmware and UEFI are infected with malware
  • Researching security of layers below the OS, such as BIOS and hypervisors
  • Why do we need to care about UEFI security?
    • There are few examples of BIOS malware
    • Infecting the BIOS requires administrator privileges
    • In most cases, administrator privileges are sufficient to achieve the goal
    • One example where this is necessary is nation-state level attacks
    • The attack itself must not be detected
    • If BIOS infection is possible, that makes it easier
    • In fact, there are cases like this at the nation-state level
    • I haven't heard of BIOS malware actually being used, but it is necessary to accumulate this knowledge
    • There is also the supply chain attack
    • When inserting attack code via the supply chain, since there is no OS, it has to be planted in the BIOS
  • How the UEFI threat works on modern PCs
    • There are major differences in platform security compared to traditional ones
      • Since Windows 11, there is a virtualization mechanism
      • HBS is the default
      • SMM privileges have also been weakened
      • UEFI malware infection is also difficult
      • S3 bootscript attacks also don't work due to SMM Lockbox
    • Most conventional methods no longer work
      • Existing SMM mechanisms don't work
      • SPI malware is also ineffective
    • Infecting the ESP has become easier
    • Bypassing HBS after reconnaissance is also important
    • Black Rota also disables HBS
    • Since stealth is emphasized, we'd like to avoid disabling HBS
    • Bypassing HBS is a prerequisite
  • What is HBS
    • VBS has two VTLs
    • Roughly, think of it as having two VMs
    • Runtime DXE modules cannot access via EPT
    • SMM modules cannot access VTL1
  • There is a method to tamper with the hypervisor during boot
    • However, there is a problem
    • It depends on the hypervisor version
    • It stops working when updated
    • It also depends on the type of hypervisor
    • The OS does not depend on the hypervisor
    • We want to bypass it without depending on hypervisor implementation or type
  • Going back to the beginning
    • Why can't Runtime DXE modules access it?
    • We need to do something about EPT paging
  • Details of paging
    • The GVA to GPA translation is in the CR3 register
    • The GPA to HPA translation is stored in the VMCS
    • The VMCS is read by vmptrld
    • Using these processors causes them to be affected
    • If we create a core where EPT is not set, we can bypass it
  • Shade Core
    • Bypasses any hypervisor-based security
    • When there are four cores, create one core with no EPT set
    • Using that core makes it work
    • Make the core to be ShadeCore unrecognized by the OS
    • Enable arbitrary code execution
  • Preventing EPT from being set
    • How is it set?
    • It is loaded from the OS loader after BIOS execution
    • EPT is set at that time
    • Tampering is done in the earlier bootloader, etc.
    • Tell the hypervisor a smaller number of cores than actual
    • Hardware information is passed to the OS loader
    • It is passed from the ACPI table
    • Change the MADT information
  • How to execute code on an unrecognized core
    • To talk to a specific core in multicore, interrupt with IPI
    • Normally it just executes the process in the interrupt table
    • How to use an arbitrary core
    • After initializing the core, initialize it with startup IPI
    • After executing the SIPI Vector, perform basic setup
    • You need to initialize that core yourself
    • Initializing means taking control of the core
  • There are two timings when initialization happens through legitimate processing
    • OS loader: ExitBootServices()
      • Since it is a BIOS function, hook it and re-initialize after it is called
    • Kernel/HV
      • Done one by one for each recognized core, not broadcast
      • With ShadeCore, since it is not recognized, this does not happen
  • Summary
    • Hook ExitBootServices()
    • The original ExitBootServices() is executed
    • Send SIPI Vector to ShadeCore
    • Jump to arbitrary code
  • How it passes
    • BIOS boot
    • OS loader
      • Execute SIPI Vector
    • OS/HV operates normally
      • ShadeCore can be used
      • Independent memory access is possible
  • How to issue commands
    • VTL0 looks at VTL1
    • It is sufficient to enable reading and writing of specific memory
    • Place magic and command in an appropriate buffer
    • ShadeCore sees this and processes it
    • However, there are also problems
      • We want to keep running the runtime DXE module, but a blue screen occurs if HVCI is present
      • ShadeCore is unaffected, but the DXE module side is affected when other cores read it
    • An EPT violation occurs
    • This is communicated through a table called MAT
    • Tamper with MAT beforehand
    • Make EPT interpret it as executable
    • MAT tampering is necessary for VTL0 and ShadeCore to communicate
    • Communication from the Kernel is also possible, and in that case MAT is not relevant
  • Demo
    • Tamper from securekernel
    • Situation with four cores
    • Connected to another PC
    • Restart
    • Wait for a while
    • After restart, cores are 4 but processors become 7
    • Start the debugger
    • Write commands and ShadeCore receives and processes them
    • Was able to attach to a process
  • Demo 2
    • Demo on Xen, not Hyper-B
    • It is good to think of VTL0 as domU and VTL1 as dom0
    • Advantage: privileges are strong from the start, but domU can also easily access
    • Tamper 3 minutes after OS boot
  • Strengths of ShadeCore
    • It can run regardless of hypervisor type
    • Existing BIOS attacks tamper after BIOS boot
    • ShadeCore doesn't tamper with components after BIOS, so it is possible
    • Since the nature of the attack method is different, it can be defined as a separate attack class
    • It can be expressed as Shading Attack
    • A method that deceives the number of cores or memory
  • Limitations of ShadeCore
    • Since the core count is different, it can be noticed there
    • Cannot support TEE
      • Encrypted with keys inside the CPU
      • ShadeCore cannot compose it
  • How to detect it
    • Verify the core count
      • Check consciously
      • It is ridiculous to always look, but constantly check the platform state
      • It is also a countermeasure against Shade-type attacks such as NIC
    • Find runtime DXE and kernel drivers used for communication
      • Without communicating with other cores, it is ineffective
  • Summary
    • SMM malware and others no longer work on modern PCs
    • Bypassing HBS is important
    • ShadeCore can bypass HBS
    • Not dependent on hypervisor type or implementation
    • Detection requires checking platform configuration information

Impressions

The attack techniques at layers below the OS were interesting content.

It would be troublesome if it were planted somewhere in the supply chain. What should we do?

Share this article

Related articles