AMD red-faced over random-number bug that kills cryptographic security

Steven

AMD red-faced over random-number bug that kills cryptographic security

Summary

AMD has confirmed a high-severity flaw (CVE-2025-62626, score 7.2) in Zen 5 processors that can undermine cryptographic randomness. The bug affects the RDSEED instruction on 16-bit and 32-bit modes: in some cases RDSEED returns 0 instead of a true random value. Because RDSEED supplies entropy used to generate strong cryptographic keys, this behaviour could be abused by an attacker with local privileges to weaken keys, potentially exposing encrypted data or credentials.

AMD is preparing microcode fixes and has already published patches for some Epyc 9005 (Turin) chips released on 28 October. Workarounds are available: use the 64-bit RDSEED where supported, hide the RDSEED feature from applications with the clearcpuid=rdseed boot parameter, or for VMs pass the -rdseed option in qemu. AMD says fixes for Ryzen and Epyc Embedded 9005 will follow later this month; other embedded and Ryzen Embedded 9000 updates are expected in January. The issue was discovered by Gregory Price from Meta and reported to the Linux kernel community.

Key Points

  • The flaw is CVE-2025-62626 (7.2) and affects RDSEED on Zen 5 CPUs running 16-bit and 32-bit modes.
  • RDSEED can return 0 instead of true entropy, risking weakened cryptographic keys used by applications.
  • An attacker needs local privileges to exploit it, so the attacker would already have substantial access.
  • Workarounds: use 64-bit RDSEED where available; add clearcpuid=rdseed to the boot cmdline; for VMs use qemu’s -rdseed option.
  • Microcode patches: Epyc 9005 (Turin) updates released on 28 Oct; Ryzen and Epyc Embedded 9005 patches due later this month; some others not until January.
  • Linux kernel maintainers have attempted mitigations (linux 6.18-rc4), though some distros reported issues entering graphical sessions after the update.
  • Root cause discovered and reported to the kernel list by Gregory Price (Meta).

Why should I read this?

Short version: if you run Zen 5 kit, this matters. Keys can be weakened if RDSEED lies to you — and while an attacker needs local access, that doesn’t make it safe to ignore. Patch where you can, or apply the simple workarounds now. Saves you from a nasty surprise later.

Author style

Punchy: this is a proper facepalm for AMD — a hardware RNG bug that talks directly to trust. Ops teams, security engineers and anyone responsible for key generation should treat this as urgent: either apply AMD’s microcode updates or enforce the mitigations. We’ve read it so you don’t have to dig through advisories.

Source

Source: https://go.theregister.com/feed/www.theregister.com/2025/11/05/amd_promises_to_fix_chips/