The TianoCore EDK II (EFI Development Kit II), the open reference implementation of UEFI used in most server motherboards, has been fixed critical vulnerability (CVE-2026-97631). The vulnerability exists in the PXE driver and can be exploited by sending specially crafted ICMP packets during network boot on IPv4 or IPv6 networks. The attack does not require authentication and allows an attacker to execute their code with UEFI firmware privileges before transferring control to the operating system, which can be used to bypass the UEFI Secure Boot verified boot mechanism and insert malicious components at the boot stage.
The problem is caused by a buffer overflow in the ICMP message handler in the UefiPxeBcDxe driver, which is used for PXE boot and receives ICMP packets during network boot to capture network errors. The overflow occurs because received error messages are placed in a fixed 504-byte buffer without checking the actual size of the received data. If more than 504 bytes are sent in an ICMP packet, the tail that does not fit is written beyond the buffer and is superimposed on the following structures.
The attack can be carried out from the global network – it is enough to know the victim’s IP address and be able to determine the moment of PXE boot. Knowing exactly what firmware is used on the victim’s server, the attacker can prepare an exploit taking into account the placement of structures in the memory of a specific firmware assembly and exploit the vulnerability to execute code at the DXE (Driver Execution Environment) level, executed in the zero protection ring.
The EDK II project was opened in 2004 by Intel and is used by all major manufacturers of firmware for server systems, data centers and cloud providers. Among others, it is used in AMI, Insyde and Phoenix firmware. The vulnerability is also present in firmware based on the Mu platform from Microsoft, which is a fork of EDK II.
The problem was identified by representatives of Anthropic and Google, and was scheduled for coordinated public disclosure in January 2027. But the information had to be made public earlier, as an engineer from the AMI company accidentally violated the embargo and published a public pull request with a fix, on the basis of which it was possible to understand the essence of the vulnerability.