Security

Langflow vulnerability: CISA reports active attacks on AI agent tool

2 min read

TL;DR Too Long; Didn’t read

The US cybersecurity agency CISA officially classified the vulnerability CVE-2026-9198 in IBM's AI platform Langflow, rated with CVSS 9.8, as actively exploited on August 4, 2026. The vulnerability allows unauthenticated attackers full remote access to standard installations, and a public exploit code has been circulating since the end of July. US federal agencies must patch by August 7, and security researchers also advise private operators to update immediately.

A masked hand reaches through an open padlock shaped like flowchart nodes towards a network cable in front of a server cabinet with IBM logo stickers and a red warning light Image generated with GPT Image 2

Key takeaways

  • CISA sets a deadline of August 7, 2026, for US federal agencies to update for the Langflow vulnerability.
  • Two chained API endpoints allow attackers full system access to standard installations without login.
  • IBM acquired Langflow in 2025 with the purchase of DataStax and integrates it into watsonx.ai.
  • A working exploit code was publicly available as early as the end of July.
  • This is the second critical remote access vulnerability confirmed by CISA in Langflow within two weeks.
  • Affected are versions 1.0.0 to 1.10.0, the update to 1.10.1 closes the vulnerability.

The US Cybersecurity Agency CISA warns of active exploitation of a critical vulnerability in Langflow, a popular open-source platform for building AI agents among developers. The vulnerability rated CVSS 9.8, CVE-2026-9198, allows attackers to gain full remote access to standard installations without login credentials. CISA has set a deadline for US federal agencies to apply the update released in July by August 7, 2026.

Two chained endpoints provide full system access

Langflow is a visual low-code platform that allows users to assemble AI agents and retrieval workflows via drag-and-drop, without deep programming knowledge. IBM acquired the tool in 2025 with the acquisition of DataStax and integrated it into its development studio watsonx.ai. The vulnerability CVE-2026-9198 chains two weaknesses in the standard configuration of the software.

The endpoint /api/v1/auto_login grants a superuser token to any network caller without requiring login. This token can then be used to access the endpoint /api/v1/validate/code, which executes submitted Python code directly via the exec() function instead of just validating it securely. In combination, attackers can gain full control over the Langflow service without knowing a single password.

According to IBM, versions 1.0.0 to 1.10.0 are affected. The update to version 1.10.1, released on July 17, completely closes the vulnerability. Additionally, by the end of July, working exploit code was circulating freely online, complete with instructions for exploitation.

CISA sets a tight deadline for federal agencies to patch

The US Cybersecurity Agency officially added the vulnerability to its Known Exploited Vulnerabilities Catalog on August 4, 2026, confirming active exploitation in the wild. Civil US federal agencies must close the vulnerability by August 7, 2026. In the same bundle, CISA also marked a vulnerability in Apache Tomcat and two weaknesses in the remote maintenance software N-central as actively attacked.

This is already the second critical remote access vulnerability that CISA has confirmed in Langflow within two weeks. Vulnerable no-code tools for AI agents are accumulating this summer. Just in June, a security researcher uncovered a comparably open network interface in the open-source platform Ruflo that allowed foreign server shells to be opened without credentials.

Those operating Langflow should immediately update to version 1.10.1 or newer and additionally isolate the auto-login endpoint from the open network via a firewall, advises IBM. While there is no legal obligation for private companies and organizations outside the US federal administration to update, security researchers also recommend immediate patching in light of the publicly circulating exploit code.

It will be crucial whether companies can keep up with the rapid growth of no-code platforms for AI agents when it comes to patching—especially since such tools often emerge outside of official IT oversight in individual departments. Whether CISA will release further details on the extent of the attacks after August 7 remains open.

Frequently asked questions

What exactly is Langflow?

Langflow is a visual low-code platform from IBM for assembling AI agents and retrieval workflows, part of watsonx.ai since the DataStax acquisition in 2025. Developers use it to build agents without extensive programming.

Has the vulnerability already been patched?

Yes, IBM released a fix with version 1.10.1 on July 17, 2026. Those running an older version between 1.0.0 and 1.10.0 remain vulnerable until the update.

Does the CISA deadline also apply to companies in Germany and the EU?

No, the directive only binds US federal agencies. However, since the exploit code is publicly circulating, security researchers recommend that all operators worldwide, regardless of location, patch immediately.

How many systems are affected worldwide?

Reliable numbers on openly accessible Langflow installations are not yet available; neither IBM nor CISA have published any information on this.

Was this the first security vulnerability in Langflow this year?

No. CISA had already classified another critical Langflow vulnerability with root access as actively exploited two weeks prior, and several other RCE and authentication vulnerabilities in the tool were reported in 2026.

Sources (5)
  1. CISA Adds Three Known Exploited Vulnerabilities to Catalog – CISA
  2. CVE-2026-9198 Detail – National Vulnerability Database
  3. IBM's agentic AI platform is under active attack - patch now – The Register
  4. CISA warns of hackers exploiting Langflow, N-central, Apache Tomcat flaws – BleepingComputer
  5. CVE-2026-9198: Langflow RCE Vulnerability – SentinelOne

Your AI update for the work week

Once a week, the most important AI news – plus one practical tip to try right away. No spam, unsubscribe anytime.

← Back to the blog