Executive Summary
In November 2026, Google Cloud disclosed three hardware-level side-channel vulnerabilities affecting AMD Turin (5th-gen EPYC / Zen 5) processors used in its C4D, N4D, H4D, Z4D, and G4 machine series, including bare-metal instances. Classified as Transient Scheduler Attacks (TSA), two flaws—TSA-SQ (CVE-2024-36350) and TSA-L1 (CVE-2024-36357)—were previously thought not to affect Zen 5; AMD has since corrected that position. A third, newly identified variant, CVE-2026-0004 (TSA-Fill), allows an attacker with low-privilege code execution on the same physical core to infer data from the L1 data cache across process or kernel boundaries. Google has mitigated inter-VM and VM-to-host attack paths at the platform level, but intra-VM exploitation requires both a stop-and-start lifecycle event and a patched guest OS kernel, which many distributions have not yet shipped to stable branches.
This disclosure underscores a broadening attack surface at the hardware abstraction layer in multi-tenant cloud environments. As workloads grow denser on shared silicon—particularly AI and confidential-compute workloads on high-core-count EPYC platforms—CPU microarchitectural vulnerabilities represent an escalating risk that neither network controls nor application-layer security alone can fully address.
Why This Matters Now
Multi-tenant cloud adoption is accelerating, and AMD EPYC Turin processors underpin a growing share of high-performance and AI workloads on Google Cloud. With CVE-2026-0004 unpatched in most stable Linux distributions and a stop-and-start lifecycle requirement that many production teams defer, the intra-VM attack surface remains open today—making timely patching and workload isolation controls immediately critical for regulated and data-sensitive industries.
Attack Path Analysis
An attacker with low-privilege code execution on an AMD Turin (Zen 5) vCPU exploits three Transient Scheduler Attack flaws (CVE-2024-36350, CVE-2024-36357, CVE-2026-0004) to speculatively infer data from co-located processes and the kernel's L1 data cache within the same VM. Without runtime microsegmentation and east-west traffic controls inside the VM boundary, the attacker pivots knowledge of leaked memory contents to escalate privileges, move laterally to co-tenanted workloads on the same physical host, and ultimately exfiltrate sensitive data through unmonitored egress channels, disrupting multi-tenant isolation guarantees across affected Google Cloud machine series (C4D, N4D, H4D, Z4D, G4).
Kill Chain Progression
This analysis maps confirmed threat intelligence to the full cloud kill chain to show where defensive gaps would emerge as an attack progresses.
Initial Compromise
Description
The attacker obtains low-privilege code execution inside a Google Cloud VM running on an AMD Turin (Zen 5) processor — achievable via a misconfigured cloud workload, exploited application vulnerability, or a malicious container deployed to a shared node. This local foothold is the prerequisite for exploiting the CPU side-channel flaws. No network-facing exploit is required; the attacker only needs the ability to execute unprivileged user-space code on a vCPU sharing a physical core.
Related CVEs
CVE-2024-36350
CVSS 5.6Transient Scheduler Attack on the Store Queue (TSA-SQ) in AMD EPYC processors including Zen 5 (Turin) allows a local attacker with low privileges to infer data from other processes or the kernel via side-channel inference on the processor store queue.
Affected Products:
AMD EPYC 9005 Series Processors (Zen 5 / Turin) – 5th Gen AMD EPYC (Turin)
AMD EPYC 7003 Series Processors (Zen 3 / Milan) – 3rd Gen AMD EPYC (Milan)
AMD EPYC 9004 Series Processors (Zen 4 / Genoa) – 4th Gen AMD EPYC (Genoa)
Exploit Status:
no public exploitCVE-2024-36357
CVSS 5.6Transient Scheduler Attack on the L1 Data Cache (TSA-L1) in AMD EPYC processors including Zen 5 (Turin) allows a local attacker with low privileges to perform side-channel inference of data residing in the L1 data cache belonging to other processes or the kernel.
Affected Products:
AMD EPYC 9005 Series Processors (Zen 5 / Turin) – 5th Gen AMD EPYC (Turin)
AMD EPYC 7003 Series Processors (Zen 3 / Milan) – 3rd Gen AMD EPYC (Milan)
AMD EPYC 9004 Series Processors (Zen 4 / Genoa) – 4th Gen AMD EPYC (Genoa)
Exploit Status:
no public exploit
MITRE ATT&CK® Techniques
Weaken Encryption
Virtualization/Sandbox Evasion: Time Based Evasion
Process Injection
Exploitation for Credential Access
Data from Local System
Escape to Host
Exploitation for Client Execution
System Information Discovery
Potential Compliance Exposure
Mapping incident impact across multiple compliance frameworks.
NIST SP 800-53 Rev 5 – Flaw Remediation
Control ID: SI-2
CISA Zero Trust Maturity Model 2.0 – Workload Runtime Monitoring and Isolation
Control ID: Workload Pillar – Visibility and Analytics
NIS2 Directive – Vulnerability Handling and Patch Management
Control ID: Article 21 – Cybersecurity Risk Management Measures
DORA (Digital Operational Resilience Act) – ICT Asset Vulnerability and Patch Management
Control ID: Article 9 – ICT Security Policies, Procedures and Tools
PCI DSS 4.0 – All System Components Are Protected from Known Vulnerabilities by Installing Applicable Security Patches
Control ID: Requirement 6.3.3
ISO/IEC 27001:2022 – Management of Technical Vulnerabilities
Control ID: Annex A 8.8
Sector Implications
Industry-specific impact of the vulnerabilities, including operational, regulatory, and cloud security risks.
Information Technology/IT
AMD Turin CPU hardware side-channel flaws (CVE-2024-36350, CVE-2024-36357, CVE-2026-0004) expose VM workloads to cross-process data leakage, requiring guest OS patches, firmware updates, and mandatory stop-start cycles across affected machine series.
Financial Services
Transient Scheduler Attacks on Google Cloud C4D/N4D/H4D bare-metal instances risk leaking sensitive financial process data across VM boundaries, threatening PCI 4.0 and NIST.800-53.SC-7 compliance without immediate Zero Trust segmentation and patching.
Health Care / Life Sciences
Cloud-hosted patient workloads on AMD EPYC Zen 5 processors face L1 cache inference attacks enabling cross-process data exposure, jeopardizing HIPAA.164.312(e)(1) and HIPAA.164.308(a)(1) compliance until guest OS mitigations and stop-start cycles are completed.
Computer/Network Security
Hardware side-channel vulnerabilities in AMD Turin CPUs undermine hypervisor isolation assumptions, exposing security vendors operating multi-tenant cloud infrastructure to privilege escalation and exfiltration risks mapped against ZTMM.Network and NIST.CSF.PR.DS-2 controls.
Sources
- Google Cloud Warns Three AMD Turin CPU Flaws Could Leak Data Across Processes in VMshttps://thehackernews.com/2026/11/google-cloud-warns-three-amd-turin-cpu.htmlVerified
- AMD Security Bulletin AMD-SB-7073: Transient Scheduler Attacks on Zen 5https://www.amd.com/en/resources/product-security/bulletin/amd-sb-7073.htmlVerified
- Xen Security Advisory XSA-525: Transient Scheduler Attacks (TSA) on AMD Zen 5https://xenbits.xen.org/xsa/advisory-525.htmlVerified
- Google Cloud Security Bulletin GCP-2026-081https://docs.cloud.google.com/support/bulletins#gcp-2026-081Verified
- AMD Technical Guidance for Mitigating Transient Scheduler Attackshttps://www.amd.com/content/dam/amd/en/documents/resources/bulletin/technical-guidance-for-mitigating-transient-scheduler-attacks.pdfVerified
- NVD Entry: CVE-2024-36350https://nvd.nist.gov/vuln/detail/CVE-2024-36350Verified
- NVD Entry: CVE-2024-36357https://nvd.nist.gov/vuln/detail/CVE-2024-36357Verified
- CISA Known Exploited Vulnerabilities Cataloghttps://www.cisa.gov/known-exploited-vulnerabilities-catalogVerified
- Linux Kernel Commit: Mitigate TSA on Zen 5https://git.kernel.org/linus/cf27412f19c9cc5909b9107cd1b3f4ab7608e362Verified
Frequently Asked Questions
Cloud Native Security Fabric Mitigations and ControlsCNSF
Based on the attack progression modeled above, these are the defensive controls that would constrain each stage.
Aviatrix Zero Trust CNSF is directly relevant because the attacker's progression from a low-privilege VM foothold toward lateral movement, credential abuse, and exfiltration depends on the absence of workload-level segmentation and controlled egress paths. By enforcing identity-aware segmentation and east-west traffic policy, CNSF would likely constrain how far extracted credentials and inferred memory contents could be operationalized across the environment.
Control: Cloud Native Security Fabric (CNSF)
Mitigation: Consistent Zero Trust policy applied at the workload boundary would likely reduce the reachability of this initial foothold to other workloads and services within the environment, constraining how broadly the compromised execution context can be observed or leveraged even before privilege escalation begins.
Control: Zero Trust Segmentation
Mitigation: Even where kernel memory leakage occurs at the CPU level, Zero Trust segmentation enforcing least-privilege identity-based policy at the workload layer would likely reduce the blast radius by limiting which resources and identities the extracted credentials could actually reach and act upon.
Control: East-West Traffic Security
Mitigation: East-west traffic controls enforcing segmentation between workloads and service identities would likely detect and constrain unauthorized lateral service-to-service connections arising from compromised credential use, reducing the attacker's ability to pivot freely across workloads even with valid-looking credentials.
Control: Multicloud Visibility & Control
Mitigation: Centralized traffic observability across cloud-native flows would likely surface anomalous outbound connection patterns from affected VM instances, reducing the attacker's ability to maintain a durable, undetected command and control channel by flagging unusual egress behavior for investigation.
Control: Egress Security & Policy Enforcement
Mitigation: Egress policy enforcement with FQDN filtering would likely constrain the attacker's ability to transfer staged data to unrecognized external destinations, reducing the probability of successful exfiltration by limiting permitted outbound paths to explicitly approved endpoints.
The residual impact would likely be constrained to the scope of the initially compromised workload, with reduced ability to monetize leaked memory contents broadly across the environment, though cryptographic material and session tokens inferred within the VM itself may still represent meaningful exposure for that specific workload's data.
Impact at a Glance
Affected Business Functions
- Cloud Virtual Machine Workload Isolation
- Multi-Tenant Data Confidentiality
- Bare-Metal and Hypervisor Security
- Google Kubernetes Engine (GKE) Node Integrity
Estimated downtime: 1 days
Estimated loss: N/A
Process-level memory data and kernel memory contents may be inferred by a co-located attacker on the same physical CPU core within a VM. In multi-tenant cloud environments running AMD Turin (Zen 5) processors, sensitive workload data such as cryptographic keys, session tokens, authentication credentials, or confidential application memory could be speculatively leaked across process boundaries. Affected Google Cloud machine series include C4D, N4D, H4D, Z4D, and G4, including bare-metal variants. No confirmed data exfiltration events have been reported; risk is theoretical but credible in shared-core cloud scenarios.
Recommended Actions
Key Takeaways & Next Steps
- • Apply guest OS vendor patches for CVE-2024-36350, CVE-2024-36357, and CVE-2026-0004 immediately upon availability, then perform a full stop-and-start (not just reboot) of all affected Turin-based VMs (C4D, N4D, H4D, Z4D, G4) to activate CPU microcode mitigations — a reboot alone is insufficient per Google's bulletin.
- • Enforce Zero Trust Segmentation with identity-based microsegmentation policies between all workloads on affected VM series to constrain lateral movement even if an attacker achieves local code execution and leaks credential material via the side-channel.
- • Deploy East-West Traffic Security controls to monitor and enforce workload-to-workload and service-to-service communication policies within VPCs hosting Turin-based instances, detecting anomalous lateral connections that may indicate use of side-channel-harvested credentials.
- • Enable Egress Security and Policy Enforcement with FQDN filtering and data-loss prevention on all outbound paths from affected VM instances to block unauthorized exfiltration of data inferred through the CPU side-channel, and activate Inline IPS (Suricata) rules to detect known memory-scraping and side-channel exploit tooling signatures.
- • Leverage Multicloud Visibility and Control to establish behavioral baselines for all workloads on affected machine series and configure alerting for anomalous outbound connection patterns, privilege escalation API calls, or unexpected cross-workload traffic flows that may indicate active exploitation of these CPU flaws.



