The vulnerability is an improper input validation flaw in the AMD ROCm Communication Collectives Library (RCCL) that permits an attacker to provide a malicious pointer, causing the library to dereference it and potentially execute arbitrary code. The weakness is classified as CWE‑822, indicating failure to properly apply defensive boundaries during data processing, which directly enables remote code execution if the library is invoked with untrusted input.

AMD Instinct™ GPUs MI210, MI250, MI250X, MI300A, MI300X, MI308X, MI325X, MI350X, and MI355X are potentially impacted. No specific version information is provided, so all current releases of these models should be considered vulnerable until an official patch notice is published.

The CVSS score of 7.7 signals a high severity risk, and the EPSS score is not available, which neither confirms nor denies the likelihood of exploitation. The vulnerability is not listed in CISA’s KEV catalog. Attackers are inferred to reach affected systems through a compromised peer rank or a network‑adjacent position that can inject invalid arguments into the RCCL library. Successful exploitation would lead to remote code execution on the host machine, allowing full control over the affected device.