Langflow CVE-2025-34291 exploitation
Summary
CISA added CVE-2025-34291 for Langflow to the Known Exploited Vulnerabilities catalog on May 21, 2026, citing evidence of active exploitation. The flaw is an origin-validation / cross-site request weakness in Langflow that can allow a malicious webpage to perform credentialed cross-origin requests, refresh tokens, and reach authenticated code-execution functionality.
The Hacker News' coverage cites earlier reporting that an Iranian state-sponsored group, MuddyWater, exploited CVE-2025-34291 to obtain initial access. Treat this as an AI-agent/workflow-platform compromise path: successful exploitation can expose Langflow workspace tokens and API keys, then cascade into downstream SaaS, cloud, model-provider, data-store, and automation integrations.
Tags
Why this matters
- Langflow-style AI workflow platforms often concentrate credentials for LLM providers, databases, SaaS APIs, vector stores, automation tools, and cloud resources.
- The exploitation path does not need a traditional phishing payload delivered to the server itself; Obsidian describes a browser-mediated chain where a victim visiting a malicious page can trigger credentialed cross-origin requests against a reachable Langflow instance.
- CISA KEV inclusion means exploitation is no longer theoretical and remediation should be prioritized like an active initial-access vector.
- Actor reporting ties at least some exploitation to MuddyWater, so exposed Langflow instances belong in Iran-nexus intrusion triage, especially where AI or automation workspaces bridge into production credentials.
Reported exploitation chain
- Langflow accepted credentialed cross-origin requests too broadly while refresh-token cookies were usable cross-site.
- The token refresh path lacked sufficient CSRF protection, letting an attacker-controlled page obtain usable authentication state through the victim browser.
- Authenticated access could then reach a Langflow endpoint that executes code by design.
- Compromise of the workspace can expose stored integration tokens, API keys, and downstream service credentials, enabling follow-on cloud/SaaS intrusion.
Defender heuristics
- Patch or upgrade Langflow per vendor guidance; CISA lists v1.9.3 as a relevant release reference for CVE-2025-34291.
- Inventory internet- and intranet-reachable Langflow deployments, especially those reachable from user browsers through VPN, ZTNA, or developer networks.
- Rotate Langflow-stored tokens after suspected exposure; do not treat application patching as enough if secrets were reachable from the workspace.
- Review Langflow logs, reverse-proxy logs, and identity-provider events for suspicious refresh-token use, unexpected cross-origin requests, new flows/components, code-validation activity, and outbound connections following browser visits.
- Hunt downstream services for use of tokens stored in Langflow around the exposure window, including cloud API calls, SaaS export activity, repository/package-registry actions, and model-provider API usage spikes.
Sources
- CISA KEV catalog entry and alert: https://www.cisa.gov/news-events/alerts/2026/05/21/cisa-adds-two-known-exploited-vulnerabilities-catalog
- CISA KEV JSON: https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json
- Obsidian Security CVE-2025-34291 analysis: https://www.obsidiansecurity.com/blog/cve-2025-34291-critical-account-takeover-and-rce-vulnerability-in-the-langflow-ai-agent-workflow-platform
- The Hacker News: https://thehackernews.com/2026/05/cisa-adds-exploited-langflow-and-trend.html