CISA Flags Actively Exploited Ray Flaw That Can Trigger Browser-Based RCE
- ID
- 15066
- Status
- summarized
- Published
- 18 Aug 2026, 2:34 PM
- Fetched
- 18 Aug 2026, 3:31 PM
- Provider
- The Hacker News
- Category
- security
- Original URL
- https://thehackernews.com/2026/08/cisa-flags-actively-exploited-ray-flaw.html
- Source URL
- https://feeds.feedburner.com/TheHackersNews
Summary
- Score
- 7.5
- Created
- 18 Aug 2026, 3:31 PM
- Tags
- Audience
- developersai_ml_learnersai_agent_users
What happened
CISA added CVE-2025-62593 (CVSS 9.4) for Ray, the open-source Python distributed computing framework for AI/ML workloads, to its Known Exploited Vulnerabilities catalog due to active exploitation. The flaw enables browser-based remote code execution via DNS rebinding—simply visiting a malicious site or ad while running Ray in a dev environment can trigger arbitrary shell code, and the browser can be used as a confused deputy to attack network-adjacent Ray instances. The fix is in Ray version 2.52.0.
Why it matters
If you run Ray for ML training or serving, upgrade to 2.52.0 immediately—especially if your Ray dashboard or API endpoints are reachable from a developer machine with a browser. The attack requires no direct network access to Ray; a developer visiting a malicious webpage is enough. Anyone running older Ray versions inside a corporate network is also at risk of lateral targeting through a compromised developer's browser.
Discussion angle
Ray's maintainers explicitly chose not to add authentication to critical endpoints like /api/jobs—what does this say about the security posture of popular open-source AI tooling, and should teams running Ray in production treat the dashboard as an untrusted attack surface by default?