Prometheus, an open-source monitoring system, has stored XSS vulnerabilities in its web UI components where metric names and label values are rendered without escaping, specifically in chart tooltips and Metric Explorer fuzzy search results. This occurs because Prometheus v3.x allows UTF-8 characters such as <, >, and " in metric and label names, which are not sanitized before being injected into innerHTML or React's dangerouslySetInnerHTML. Attackers who can inject metrics through compromised scrape targets, remote write, or OTLP receiver endpoints can execute arbitrary JavaScript in the browsers of users viewing these metrics. This can lead to actions such as configuration exfiltration, data deletion, or Prometheus shutdown depending on enabled flags. The vulnerability is fixed in Prometheus versions 3.5.2 and 3.11.2. Workarounds include restricting exposure of remote write and OTLP receivers, ensuring scrape targets are trusted, disabling admin or mutating API endpoints in untrusted environments, and avoiding clicking untrusted links that may generate malicious label names.

Exploitation allows arbitrary JavaScript execution in the browser of any Prometheus user who views the maliciously crafted metric or label in the web UI. This can lead to sensitive configuration data exfiltration, deletion of monitoring data, or shutdown of the Prometheus server depending on enabled features. The vulnerability requires the attacker to inject metrics via compromised or attacker-controlled scrape targets or remote write/OTLP endpoints.

A fix is available in Prometheus versions 3.5.2 and 3.11.2; upgrading to these or later versions is recommended. Until updates can be applied, restrict exposure of the remote write receiver (--web.enable-remote-write-receiver) and OTLP receiver (--web.enable-otlp-receiver) to trusted sources only. Ensure all scrape targets are trusted and not attacker-controlled. Avoid enabling admin or mutating API endpoints such as --web.enable-admin-api or --web.enable-lifecycle in environments where untrusted data may be ingested. Users should also avoid clicking untrusted links that may contain functions like label_replace, which can generate malicious label names and values.