The distributed computing framework used to train models is actively exploited; the vulnerability is rated 9.4 out of 10 and fixed in version 2.52.0.
On August 17, 2026, CISA added CVE-2025-62593 to its catalog of actively exploited flaws, with a remediation deadline of August 20 for U.S. federal civil agencies — just three days. The flaw affects Ray, the widely used open source distributed computing framework for model training, data processing, and large-scale Python workloads.
The flaw stems from insufficient protection of Ray's HTTP entry points, including those used to submit tasks — and therefore execute code. The safeguard was based on a cursory check of the User-Agent header, which was supposed to start with "Mozilla." Combined with a DNS rebinding attack, this weakness allows an attacker to execute arbitrary code on the workstation of a developer who is running Ray locally, at the sole cost of a visit to a booby-trapped site or the display of a malicious advertisement. The Firefox and Safari browsers are particularly exposed.
The publisher rates the vulnerability 9.4 out of 10 according to the CVSS 4.0 scale. The fix is available in Ray 2.52.0; all previous versions are affected.
Read the full article on the original source
Go to the source →