CISA Adds Actively Exploited Flaws in Langflow, Tomcat and N-central to Its Catalog

On August 5, 2026, the U.S. Cybersecurity and Infrastructure Security Agency (CISA) added three new entries to its Known Exploited Vulnerabilities (KEV) catalog. The basis for the move was evidence that all of them are already being used by threat actors.

Three Vulnerabilities

The first — CVE-2026-9198 (CVSS 9.8) — is a code injection flaw in the Langflow platform. It allows an unauthenticated attacker to achieve full remote code execution in the default configuration. The issue was already fixed back in July 2026 in version 1.10.1. Exactly how it is being exploited remains unknown, but flaws in this open-source tool for building AI applications have been targeted repeatedly in recent months.

The second — CVE-2026-34486 (CVSS 7.5) — affects Apache Tomcat: the lack of encryption of sensitive data makes it possible to bypass the EncryptInterceptor component responsible for protecting messages between cluster nodes. Fixes were released in April 2026 (versions 11.0.21, 10.1.54 and 9.0.117).

The third — CVE-2026-18556 (CVSS 8.2) — is an authentication bypass in N-able N-central. The initial patch turned out to be incomplete, forcing the vendor to release a new one, tracked as CVE-2026-18577. The latter made it into the catalog as early as Monday, and it is now clear that both are being exploited.

Autonomous AI in the Service of Attackers

The most intriguing part concerns the Tomcat vulnerability. According to Palo Alto Networks' Unit 42, its exploitation is attributed to a Russian-speaking — or more precisely Chinese-speaking — group operating under the aliases knaithe and KnYuan out of the city of Zhuhai. The attacker used the DeepSeek model via the Hermes Agent framework in the role of an “offensive operator” against internet-facing devices.

When the first attempt to breach Langflow (CVE-2026-33017) failed because of the target's strict configuration, the AI agent independently carried out reconnaissance and found other, more valuable vulnerabilities — including in n8n. In parallel, the group manually attacked known flaws in Citrix NetScaler, Marimo, Tomcat and IKE VPN.

In total, more than 460 targets were hit — using both autonomous and manual methods. According to the researchers, the agent effectively allowed DeepSeek to narrow down the pool of targets, likely in order to save computing resources. In a matter of minutes, the system performed hundreds of hours' worth of target analysis, while managing its own computing power along the way.

U.S. federal civilian agencies have been ordered to apply the fixes by August 7, 2026.

Source: The Hacker News