A critical vulnerability in TianoCore EDK II, the open source codebase hardware makers use to build UEFI firmware, allows an unauthenticated network attacker to corrupt memory and potentially execute code before the operating system loads. The flaw, tracked as CVE-2026-97631, is rated 9.5 out of 10 and sits in the PXE network boot driver, in the handling of ICMP error messages over both IPv4 and IPv6.
The mechanism is a bounds failure. During network boot, the driver copies incoming ICMP error messages into a fixed 504 byte field, and the copy logic can write past it, with the destination pointer advancing by cumulative length rather than the current fragment’s length. An attacker who can deliver a unicast ICMP error message to a machine while it is actively PXE booting needs no authentication and no user interaction. The outcomes range from deterministic corruption that hangs the boot process to, for an attacker who knows the target firmware build, execution of supplied instructions in the Driver Execution Environment phase, before Secure Boot and operating system controls exist.
The disclosure was not orderly. The flaw was discovered by Alex Matrosov of Anthropic and Dan Holmes of Google, with a coordinated disclosure planned for January. AMI, a firmware vendor, published a pull request describing the flaw and including a fix ahead of that schedule, which prompted the public advisory. TianoCore merged corrective changes through an upstream pull request on October 15, 2026.
That upstream fix is not the end of the exposure. Virtually all UEFI suppliers, including AMI, Insyde and Phoenix, build on the EDK II reference code, and Microsoft’s hardened EDK2 fork, Project Mu, also used the vulnerable component. Each vendor must still ship its own updated firmware, and each operator must still apply it, which is why a flaw in a reference codebase threatens most servers rather than one vendor’s machines.