CVE-2026-97631 in TianoCore EDK II lets attackers run pre-boot code

Alex Matrosov of Anthropic and Dan Holmes of Google are the researchers who found CVE-2026-97631, a memory-corruption flaw in the ICMP component of TianoCore EDK II, and the flaw was disclosed publicly Oct. 27, 2026; attackers who exploit it can run malicious code before the system starts.

The malicious code runs during early boot, before the operating system or the firmware can switch on defenses such as Secure Boot. Secure Boot normally establishes a chain of trust for every code component, including the Trusted Platform Module, as each component loads during boot. The initial payload runs only once, lacks persistence, and opens the door to others, but attackers can use that one-time execution to add payloads that run early each time the server boots. The flaw bypasses that chain of trust.

TianoCore EDK II is open source code that hardware makers and cloud providers use when they build UEFI firmware for boot-up, and it serves as the reference codebase for that firmware. Nearly all UEFI suppliers, also called independent BIOS vendors and including AMI, Insyde, and Phoenix, rely heavily on the code, with virtually all suppliers using the reference code extensively in their products.

In 2025, a UEFI Forum representative called EDK2 the top implementation of the UEFI specifications. Microsoft’s hardened EDK2 fork, Project Mu, used the code containing the vulnerable component, and most server vendors and data centers around the world run this affected boot-time firmware on millions of devices.

Matrosov and Holmes wanted the flaw kept out of public view until January, with no year attached to the coordinated disclosure. Hardware and motherboard maker AMI published a pull request in September 2026 that described the flaw and included a code fix, ahead of the public advisory. TianoCore Project members issued an advisory in October 2026 after spotting the pull request.

Applying patches for UEFI flaws across millions of devices made by hardware makers and cloud providers is slow and difficult, and fixing these flaws takes cumbersome, time-consuming work across the supply chain. Since the patch code is public, skilled hackers can easily learn how to trigger the flaw and build working exploits, which lowers the barrier for exploit development across the supply chain. Because the code runs before operating-system defenses load, any pre-boot compromise can sit beneath those defenses. This announcement came six days after the flawed disclosure of a separate critical code-execution bug in PHP.