CVE-2023-20591 is a medium severity flaw (CVSS 6.5) in AMD EPYC firmware. The IOMMU is not properly re-initialized during a DRTM event, so an untrusted platform configuration can persist. AMD fixes it through updated Platform Initialization (PI) firmware, delivered as a BIOS update from your server vendor.
What's affected
- Product named in the record: AMD EPYC 8024P and 8024PN firmware and hardware (the 8004 series). The record lists no firmware version ranges.
- AMD bulletin AMD-SB-3003 (August 2024) marks CVE-2023-20591 as "Not affected" for 1st Gen (Naples) and 2nd Gen (Rome) EPYC.
- For 3rd Gen (Milan) the bulletin lists MilanPI 1.0.0.B (2023-06-08) as the minimum firmware that mitigates it. For 4th Gen (Genoa) it lists Genoa 1.0.0.8 (2023-06-09).
- Severity: CVSS 3.1 score 6.5 Medium, vector AV:N/AC:H/PR:N/UI:N/S:C/C:L/I:L/A:L.
The record and the bulletin do not line up exactly. The record names the EPYC 8024P and 8024PN parts, while the bulletin text we read lists mitigation versions for Milan and Genoa generations. Read the bulletin table for the 8004 series directly before you decide a given host is clear.
How to check if you're exposed
This is a firmware issue, so a package scanner will not see it. Find the platform firmware version on each host and compare it with the table in AMD-SB-3003.
sudo dmidecode -s bios-version
sudo dmidecode -s processor-version
The first command prints the BIOS version your server vendor shipped. The second prints the processor model, which tells you which EPYC generation you are looking at. Your hardware vendor maps its BIOS releases to AMD PI versions, so look in its release notes for the PI version inside each BIOS.
What to do now
- Identify every AMD EPYC host, and note the processor model and BIOS version.
- Open AMD-SB-3003 and find your processor generation. Confirm whether CVE-2023-20591 is listed as affected and which minimum PI firmware applies.
- Ask your server vendor for a BIOS release that includes that PI version or newer, and schedule the flash in a maintenance window.
- Prioritize hosts that run hypervisors and use DRTM based launch features, since the weakness concerns hypervisor memory.
- Record the BIOS version after the update so the next audit can confirm it.
Why it matters
The weakness is improper initialization (CWE-665 (Improper Initialization)). During a dynamic root of trust (DRTM) event, the IOMMU should be reset to a known state. Here it is not fully re-initialized, so a configuration set before the event may survive. Per the description, that could let an attacker read or modify hypervisor memory.
The CVSS vector rates attack complexity high and shows a changed scope, which fits a flaw that sits below the operating system and affects things above it. For most fleets the practical risk is low, but firmware rarely gets patched on its own. It needs a plan and an owner. CISA has not listed this CVE in its Known Exploited Vulnerabilities catalog, according to the record we hold, and none of the pages we read said it is being exploited.
Quick check
Will a vulnerability scanner report this?
Only if it reads firmware inventory. Package dependency scans, including the free scan at gold.safeguard.sh/scan, look at libraries in your manifests, not BIOS versions. Use your hardware inventory for this one.
Is there a workaround besides updating?
For the generations whose tables we read, AMD says the PI firmware is the mitigation, and for Rome it says there are no other options. Check the bulletin for your generation.
Sources
Last reviewed: 19 October 2026.