November 17, 2026

Google Cloud Issues Security Update for AMD Turin-Based VMs Following Discovery of New CPU Side-Channel Vulnerabilities

In a recent security advisory that highlights the persistent challenges of securing modern processor architectures, Google Cloud has notified customers that Virtual Machines (VMs) powered by the latest AMD "Turin" chips require urgent attention. To protect against a trio of CPU side-channel vulnerabilities, administrators are being instructed to apply guest operating system patches and perform a full "stop-and-start" cycle of their instances.

The vulnerabilities, which involve complex architectural flaws in how processors handle transient data, affect several of Google Cloud’s high-performance machine series. While Google has already implemented broad mitigations at the hypervisor level to protect the integrity of the cloud environment itself, the responsibility for securing the internal environment of the VM—protecting one process from another—now rests with the end users.

The Scope of the Turin Vulnerabilities

The hardware at the center of this disclosure is the fifth-generation AMD EPYC processor, developed under the code name "Turin." These processors utilize the Zen 5 microarchitecture, which was designed to offer significant performance gains and efficiency improvements over previous generations. However, recent security research has revealed that even these cutting-edge chips are susceptible to a class of flaws known as Transient Scheduler Attacks (TSA).

Google has identified several specific machine series within its cloud portfolio that are impacted by these findings. These include the C4D, N4D, H4D, Z4D, and G4 series. The scope of the vulnerability extends to bare-metal instances of the C4D and Z4D series, where the lack of a traditional virtualization layer means the hardware-level flaws must be addressed directly through firmware and operating system updates.

The three specific vulnerabilities in question are categorized as side-channel attacks. These do not involve traditional software bugs or "buffer overflows" in the traditional sense. Instead, they exploit the fundamental way modern CPUs optimize performance through speculative execution and task scheduling. By observing the timing or state changes in the processor’s internal components, an attacker with the ability to run code on the same physical core as a target process could potentially infer sensitive data.

Understanding Transient Scheduler Attacks

Transient Scheduler Attacks represent a sophisticated evolution of the side-channel vulnerabilities that have plagued the semiconductor industry since the discovery of Spectre and Meltdown years ago. These attacks target the "transient" state of a processor—data that is processed during speculative execution but is supposed to be discarded if the processor determines the execution path was unnecessary.

The first two vulnerabilities, TSA-SQ (tracked as CVE-2024-36350) and TSA-L1 (tracked as CVE-2024-36357), are not entirely new to the security community. They were first disclosed in July 2025 and were initially thought to primarily affect older AMD EPYC architectures, such as Rome, Milan, and Genoa. At the time of the original disclosure, there was a prevailing belief that the Zen 5 architecture might be inherently immune to these specific attack vectors.

Google Cloud Warns Three AMD Turin CPU Flaws Could Leak Data Across Processes in VMs

However, recent analysis has forced a correction of that narrative. AMD has now clarified that its Zen 5 processors are indeed impacted, albeit in a manner that differs from previously characterized AMD silicon. This "different" impact necessitated a re-evaluation of the security posture for the Turin chips, leading to the current round of warnings from major cloud providers like Google.