Critical EDK II Flaw Allows Pre-Boot Code Execution on Most Servers Worldwide

A critical severity vulnerability has been discovered in EDK II (EFI Development Kit 2), the world's most widely used firmware development environment for UEFI platforms. Tracked as CVE-2026-97631 and rated 9.5 out of 10 on the CVSS 4.0 scale — nearly the maximum possible score — the flaw resides in the PXE Base Code driver (NetworkPkg/UefiPxeBcDxe) and can allow an unauthenticated attacker to execute arbitrary code during the pre-boot phase, before Secure Boot or measured boot has verified any OS components.

How the flaw works

The problem lies in the ICMP (Internet Control Message Protocol) error handlers in both the IPv4 and IPv6 handlers of the network boot driver. During network boot, the driver keeps a receive request open for ICMP error messages and copies each received message into a fixed 504-byte field called `Mode.IcmpError` in its private context structure.

The copy loop has a bounds handling flaw in both IPv4 and IPv6 handlers: for messages longer than 504 bytes, the code continues copying bytes past the field, advancing the destination pointer by cumulative rather than per-fragment length. Nothing in the receive path earlier restricts the message size. This sets up a classic buffer overflow (CWE-120) combined with incorrect buffer size calculation (CWE-131) and out-of-bounds write (CWE-787).

The result is twofold: deterministic corruption of the driver's private data, causing a CPU exception inside the DXE core on the next ICMP error (boot hang, denial of service), and the possibility, for an attacker who knows the target's firmware configuration, to inject a counterfeit event in the same packet and have `CoreSignalEvent()` dispatch attacker-controlled bytes in the DXE phase — ring 0, before any OS security control exists.

Who is affected

EDK II is maintained by the TianoCore foundation and is the de-facto reference implementation for the UEFI specification. Vendors include American Megatrends (AMI), Intel, Insyde Software, Phoenix Technologies, and Microsoft. Machines using EDK II that network-boot via PXE — and, crucially, are configured to use IPv6 — are all vulnerable to exploitation.

Intel told Ars Technica it knows of no commercial products currently under security support that are affected, but noted that OEMs or system makers might integrate the vulnerable functionality.

The exact scope of affected products remains unclear, and several companies contacted by Ars Technica had not responded at publication. Systems that boot exclusively from disk are not affected — the exposure requires access to the relevant network during the PXE boot sequence.

Discovery and patching

The vulnerability was discovered by Alex Matrosov of Anthropic and Dan Holmes-Mitra of Google, who also provided the recommended fix. Disclosure came earlier than planned after motherboard maker AMI published a pull request containing the fix details — which, ironically, makes it easier for advanced attackers who wish to learn how the vulnerability is triggered.

The pull request (#13302) adds size validation logic before copying and includes unit tests for the conditions the fix resolves. The affected version of EDK II is master as of 2026-09-14. The latest stable release, `edk2-stable202608` (August 21, 2026), was also listed as affected.

What to watch for

With publicly available patch code, it will be trivial for advanced hackers to learn how the vulnerability is triggered and build working exploits. The exposure window is the duration of PXE boot — which means the risk is concentrated at specific moments, but the attack surface is broad in datacenter environments where network boot is part of the normal provisioning workflow.

What makes this vulnerability particularly dangerous is that it operates at a level below any OS protection. A successful exploit would compromise the entire platform, including the ability to modify or implant malware that survives OS reinstalls. The analogy with the BlackLotus case — the first publicly documented UEFI bootkit that defeated Secure Boot on a stock Windows 11 machine — illustrates the type of scenario infrastructure administrators want to avoid.

The question is whether the open-source firmware community, which has worked diligently with the coordinated disclosure process, will be able to produce OEM-distribution fixes in time, or whether hundreds of thousands of servers will remain at risk until firmware updates reach datacenters.

Sources: TianoCore EDK II Security Advisory, Linxi News, CVE Record

✓ Independent sources cross-checked and verified before publishing