GHSA-g9ww-x58f-9g6m — contrast
HIGHGHSA-g9ww-x58f-9g6m is a high-severity (CVSS 8.4) remote code execution vulnerability in github.com/edgelesssys/contrast. A fix is available for github.com/edgelesssys/contrast — see the affected versions and patch details below.
Contrast BadAML injection allows arbitrary code execution
Real-World Exposure
github.com/edgelesssys/contrastReal-time download stats are indexed for npm and PyPI packages. This vulnerability affects Go packages — download data is not available via public APIs for these ecosystems.
Description
BadAML
BadAML is an AML injection attack that exploits the ACPI interface and allows arbitrary code execution in a confidential VM. The attack was first published in 2024:
- https://blackhat.com/eu-24/briefings/schedule/#aml-injection-attacks-on-confidential-vms-42723
- https://dl.acm.org/doi/pdf/10.1145/3719027.3765123
Impact
An attacker with control over the host (which is assumed in the attacker model of Contrast) can execute malicious AML code to gain arbitrary code execution within the confidential guest.
AML is byte code embedded in ACPI tables that are passed from the host (QEMU) to the guest firmware (OVMF), and then passed from OVMF to the Linux kernel. The Linux kernel has an interpreter that executes the AML code. An attacker can craft a table with malicious AML code and the kernel will execute it. AML is Turing-complete and the interpreter has access to the full guest memory, including private pages.
See the paper for a detailed description and background of the attack.
Note that this is not a vulnerability specific to Contrast, but rather a generic vulnerability in Confidential Computing setups that use the ACPI interface.
Affected platforms
This issue affects the SNP platforms supported by Contrast: Metal-QEMU-SNP and Metal-QEMU-SNP-GPU.
Users on these platforms should switch to the fixed Contrast version immediately.
Metal-QEMU-TDX isn't affected, as the content of the ACPI tables is covered by the runtime measurements (measured into RTMR 0 by OVMF) on Intel TDX.
Patches
A sandbox similar to the one proposed in the paper has been implemented in the Linux kernel used by Contrast. The sandbox denies access to private memory pages by doing a page table lookup on every read/write by the AML interpreter.
This mitigates the attack completely: While an attacker can still run AML code, the code cannot read or modify private memory pages. Shared pages are readable/writable by the host hypervisor anyway.
Affected Packages
| Ecosystem | Package | Vulnerable range | Fix |
|---|---|---|---|
| 🐹Go | github.com/edgelesssys/contrast | all versions | 1.18.0go get github.com/edgelesssys/contrast@v1.18.0 |
Detection & mitigation playbook
Open-source dependencyDetect
Scan your dependency tree (package-lock.json, pnpm-lock.yaml, requirements.txt, go.sum, etc.) for github.com/edgelesssys/contrast, including transitive dependencies — a direct dependency you never call can still pull in a vulnerable version.
Fix
Update github.com/edgelesssys/contrast to 1.18.0 or later, then make sure no transitive (indirect) dependency still pins the vulnerable range — O3 confirms GHSA-g9ww-x58f-9g6m is resolved across your whole dependency graph.
Workarounds
If you can't upgrade right away: gate or disable the affected feature, validate untrusted input at the boundary, and avoid passing attacker-controlled data into the vulnerable path. O3's runtime protection blocks exploitation in production as an interim safeguard until the upgrade lands.
How O3 protects you
O3 Security's impact-aware SCA analyses which vulnerable code paths your application actually calls, so a match like GHSA-g9ww-x58f-9g6m can be triaged on real exposure rather than presence alone.
Tailored to GHSA-g9ww-x58f-9g6m. Runtime protection reduces exposure until a permanent patch is applied and verified — it complements patching, it doesn't replace it.
Frequently Asked Questions
Is GHSA-g9ww-x58f-9g6m in your dependencies?
O3 Security finds GHSA-g9ww-x58f-9g6m across Go dependencies, including transitive ones, and its impact-aware SCA ranks findings by whether your code actually calls the vulnerable path.