thehackernews.com web signal

Langflow RCE Flaw Exposes 7,000 AI Instances

cybersecurity agents ai-security vulnerability ai-agents

Key insights

  • CVE-2026-5027 exploits a filename sanitization failure in Langflow's file upload endpoint, enabling arbitrary filesystem writes without authentication.
  • Langflow's default unauthenticated auto-login means roughly 7,000 publicly exposed instances are reachable by attackers with no credentials required.
  • Tenable disclosed publicly on March 27, 2026 after three failed contact attempts in January-February 2026, and no patch has been released.

Why this matters

Langflow is a widely used open-source platform for building AI agent applications, so a working unauthenticated RCE against roughly 7,000 publicly exposed instances means attackers can execute arbitrary code on AI development infrastructure with no patch in sight. The three-month gap between Tenable's public disclosure on March 27, 2026 and no available fix as of June 2026 reveals a maintainer-response failure that extends confirmed active exploitation indefinitely. MuddyWater, an Iranian state-sponsored group, already weaponized a related Langflow vulnerability (CVE-2025-34291), confirming the platform is actively targeted by state-sponsored actors beyond opportunistic scanning.

Summary

An unpatched path traversal flaw in Langflow is actively exploited against roughly 7,000 publicly exposed instances, most in North America. CVE-2026-5027 (CVSS 8.8) targets the POST /api/v2/files endpoint, which fails to sanitize filenames in multipart form data, letting attackers write files anywhere on the filesystem via path traversal. Langflow's default auto-login removes all authentication barriers, giving unauthenticated attackers direct access and enabling remote code execution. Essentially: (Langflow, Tenable) -- high-severity RCE, no patch available. - Tenable made three contact attempts in January-February 2026; public disclosure came March 27, 2026. - Current attacks write test files to victim systems, consistent with active reconnaissance before deeper exploitation. - MuddyWater, an Iranian state-sponsored group, already weaponized a related flaw, CVE-2025-34291. No fix is available; the 7,000 exposed instances depend entirely on manual hardening.

Potential risks and opportunities

Risks

  • Organizations running Langflow for AI agent development have no patch available as of June 2026, leaving roughly 7,000 reachable instances exposed to unauthenticated RCE indefinitely.
  • MuddyWater, which already weaponized CVE-2025-34291, now has a clear path to exploit CVE-2026-5027 for deeper persistence within previously targeted networks.
  • Active attacker reconnaissance via test-file writes could escalate to full RCE payloads across the 7,000 North America-heavy exposed instances before any patch ships.

Opportunities

  • Tenable, having discovered and publicly disclosed CVE-2026-5027, is positioned to expand its AI dev toolchain security practice as enterprise demand for AI-specific vulnerability research grows.
  • Security teams can use this incident as direct justification for network segmentation policies that isolate AI development platforms from production infrastructure.
  • Managed AI workflow platforms that enforce authentication by default gain concrete competitive differentiation against self-hosted Langflow in enterprise security evaluations.

What we don't know yet

  • Whether Langflow maintainers have committed to a patch timeline as of June 2026, more than two months after public disclosure on March 27, 2026
  • What capabilities attackers are establishing beyond the test-file writes Tenable observed, and whether any confirmed intrusions have been attributed to CVE-2026-5027 specifically
  • Whether the approximately 7,000 exposed instance count includes cloud-hosted Langflow deployments or only self-hosted instances accessible via public IP