The headliner is CVE-2026-9198, a 9.8-severity code-injection bug in Langflow that hands an unauthenticated attacker full remote code execution on a default deployment. It chains two mistakes: an endpoint that mints superuser bearer tokens to any caller, and a code-validation endpoint that will happily run arbitrary Python. IBM patched it in Langflow OSS 1.10.1 back on July 17, but default installs stayed exposed, and now it's being exploited in the wild. Alongside it, CISA flagged CVE-2026-34486 — an EncryptInterceptor bypass in Apache Tomcat's clustering — and another N-able N-central authentication bypass. The Tomcat exploitation has a detail worth pausing on: it's been attributed to an AI-driven autonomous campaign run by a Zhuhai-based actor using DeepSeek through the Hermes Agent framework as the offensive operator. Federal civilian agencies have until August 7 to remediate under the KEV mandate, but the deadline is a useful forcing function for everyone. If you run Langflow anywhere reachable, patch to 1.10.1 or take it off the network today; the token-minting endpoint means there's no authentication step to slow an attacker down.