Canadian Cyber Security Journal
SOCIAL:
Filed under: Featured, TechTalk

CISA Flags Actively Exploited Ray Flaw (CVE-2025-62593): What Canadian AI Teams Need to Know

What Happened

CISA added CVE-2025-62593, a critical code injection vulnerability in the open-source Ray distributed compute framework, to its Known Exploited Vulnerabilities catalog on August 17 after confirming active exploitation. The CVSS 9.4 flaw sits in Ray’s handling of HTTP API endpoints, including /api/jobs and /api/job_agent/jobs, allowing an attacker to submit or trigger arbitrary code execution against a vulnerable local Ray instance. Researchers say the vulnerability supports browser-based remote code execution too, using a DNS rebinding attack launched from Firefox or Safari against Ray instances reachable from a victim’s local network. Operators of the RondoDox DDoS botnet incorporated the flaw into their toolset before public disclosure, and separate campaigns have used exposed Ray instances for cryptocurrency mining. The fix ships in Ray 2.52.0. Federal civilian agencies in the US face an August 20 remediation deadline under CISA’s binding directive.

Why This Matters for Canadian Organizations

Ray runs distributed training and inference infrastructure behind a growing share of Canadian AI development, from research labs to enterprise machine learning teams building on the framework’s scaling features. An unauthenticated attacker only needs network reach to a Ray cluster’s dashboard port to execute code, and DNS rebinding turns an ordinary employee visiting a malicious webpage into an entry point even when a cluster sits behind what a team assumed was a private network. Organizations subject to OSFI Guideline B-13 face expectations around timely remediation of critical, actively exploited vulnerabilities, and this flaw meets the threshold directly given confirmed botnet weaponization. Any Ray deployment holding proprietary models, training data, or customer information tied to Canadian operations carries exposure under PIPEDA if compromise leads to unauthorized access to personal information.

What to Do

Security teams should inventory every Ray deployment, confirm dashboard and job-submission endpoints sit off the public internet, and upgrade to Ray 2.52.0 without delay. Where upgrading takes time, restrict access to Ray’s API ports through network segmentation and block browser access to the dashboard from untrusted networks. Full technical detail is available from The Hacker News and CISA.

Enjoy this article? Don’t forget to share.