On August 12, 2026, it emerged that the Trusted Platform Module 2.0 firmware (fTPM 2.0)—the integrated security co-processor in AMD and Intel processors—contains two critical vulnerabilities that undermine the very foundation of the system trust chain. CVE-2026-6726 and CVE-2026-6727, with CVSS scores of 8.5 and 8.3 respectively according to heise online, allow an attacker with privileged local access to extract cryptographic secrets and forge TPM keys. The significance extends beyond the technical: for the first time, a flaw in the Trusted Computing Group's reference code has created a cross-vendor problem, forcing the two historic rivals into unprecedented coordination.
- The Trusted Computing Group's fTPM 2.0 reference code, integrated by AMD and Intel, contains two vulnerabilities: CVE-2026-6726 (CVSS 8.5) and CVE-2026-6727 (CVSS 8.3).
- CVE-2026-6727 is a timing side-channel attack on RSA-OAEP that extracts secrets by measuring co-processor response times; CVE-2026-6726 exploits the failure to invalidate discarded keys to plant forged keys at the same memory address.
- Exploitation requires privileged local access: the primary impact falls on enterprise devices, not consumer endpoints.
- AMD confirmed all Zen architectures from Ryzen 1000 through Ryzen 9000 and Ryzen AI 400 are affected; Intel coverage extends to Core Ultra 200 and third-generation Xeon Scalable (Ice Lake). BIOS updates will roll out over the coming weeks and months.
The Mechanism of Collapse: Timing Attacks and Ghost Keys
The first flaw, CVE-2026-6727, is a classic timing side-channel with an unusual target: the RSA-OAEP implementation in fTPM 2.0. As heise online describes, "specially crafted software sends countless commands to the co-processor and measures the time the system takes to respond." The program fills the RSA-OAEP encryption blocks with correct and incorrect padding values; the response times differ, enabling gradual information extraction. The CVE-2026-6726 record, verified on cve.org, describes an information leak in the TCG TPM 2.0 reference code: "a local attacker with elevated privileges can obtain credentials from a TPM-aware CA for a forged TPM key."
The second flaw, CVE-2026-6726, strikes at the TPM key lifecycle. As heise online reports, "under certain circumstances, a TPM does not properly invalidate discarded keys, allowing attackers to present a new forged key at the same address as a genuine key." This "replacement attack" compromises upstream cryptographic verification, opening the door to deep system manipulation.
"The Trusted Platform Module is the most important link in the trust chain of PCs. Ironically, this TPM is vulnerable." — heise online
A Shared Architecture Problem: The Role of the TCG
The common denominator is the fTPM 2.0 reference code supplied by the Trusted Computing Group, the standards body that defines TPM specifications. AMD and Intel integrated this code into their respective BIOS implementations, propagating the same vulnerability across two normally isolated ecosystems. The cross-vendor scope makes coordination unavoidable: a patch for Debian's libtpms is insufficient if the microcontroller firmware remains exposed.
Microsoft has released an advisory for CVE-2026-6727 with a full CVSS 3.1 vector: AV:L/AC:H/PR:N/UI:N/S:C/C:H/I:N/A:N/E:U/RL:O/RC:C, rated "Important." Redmond has deployed patch KB5123303 for Windows Server 2022. Heise online also notes software updates for the Debian libtpms library and "numerous Windows versions," indicating the software layer above the firmware also requires remediation.
The Attack Surface: Who Is Truly at Risk
The risk profile is heavily skewed toward the enterprise. Heise online is explicit: "local access with privileged rights is required for exploitation. This makes the vulnerabilities of little relevance for private users. They are relevant, for example, in enterprises: attackers could compromise a notebook, present it as a secure company device, and thus infiltrate the entire corporate network." The TPM serves as the root of trust for BitLocker, UEFI secure boot, and device authentication in managed environments.
AMD has listed affected platforms: all desktop, notebook, and embedded processors with Zen architecture, from Ryzen 1000 through Ryzen 9000 and Ryzen AI 400. For servers, only the Epyc 4004 and 4005 series are listed as vulnerable; the 7000 and 9000 series do not appear, with heise online noting this omission without confirming the reason. Intel covers desktop, notebook, and embedded processors up to the Core Ultra 200 series, and servers up to the third-generation Xeon Scalable (Ice Lake) from 2021.
What to Do Now
Priority actions for IT and security teams derive directly from verified sources:
- Monitor availability of BIOS updates distributed by AMD and Intel to system manufacturers: rollouts are expected in the coming weeks and months, with timelines varying by vendor.
- Apply software patches already released: KB5123303 for Windows Server 2022 is available, along with updates for the Debian libtpms library and compatible Windows versions.
- Assess impact on fleets with AMD Ryzen 1000-9000 and Intel Core up to Ultra 200 processors, verifying patch status of corporate mobile devices that are off-site or not centrally managed.
- Review privileged local access policies on critical devices: exploitation requires this prerequisite, which remains the only controllable element until corrected firmware arrives.
The Hidden Inconsistency: Two Scores, One Source
A limitation in the dossier warrants transparency: the CVSS 8.5 score for CVE-2026-6726 comes exclusively from heise online, while the official record on cve.org does not include a score in the available dump. For CVE-2026-6727, Microsoft and heise converge on the data. The CVE-2026-6726 record describes the information leak mechanism but does not independently quantify severity relative to the journalistic source. Severity remains high in both cases, but the difference in primary documentation is an element technical readers will rightly appreciate.
Why This Flaw Changes the Concept of Hardware Security
The TPM is designed to be the hardest link in the chain—the component that cannot be compromised even if the operating system falls. Discovering that this link shares its code with a competitor—and that both inherit the same flaw—raises an architectural question deeper than the vulnerability itself. The Trusted Computing Group standardized a root of trust that, for all its physical isolation, was united by software. The consequence is not merely technical: it is a lesson on the risks of cryptographic monoculture, even when that monoculture wears two different brands.
Sources
- https://www.heise.de/en/news/The-security-co-processor-in-many-CPUs-is-insecure-11411956.html
- https://www.amd.com/en/resources/product-security/bulletin/amd-sb-7039.html
- https://www.amd.com/en/resources/product-security/bulletin/amd-sb-3019.html
- https://www.cve.org/CVERecord?id=CVE-2026-6726
- https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-6727
- https://www.amd.com/en/products/adaptive-socs-and-fpgas/technologies/design-security.html
- https://www.heise.de/plus/abo?affiliateId=30165&wt_mc=intern.abo.plus.hp_ea_2025.ho_navilink.desktop.desktop
- https://www.heise.de/sso/registration?forward=https%3A%2F%2Fwww.heise.de%2Fen%2Fnews%2FThe-security-co-processor-in-many-CPUs-is-insecure-11411956.html
Information verified against cited sources and current as of publication.