CVE-2026-21445

Langflow is like a “workshop” for building and running AI agent workflows, including storing user messages and related data. This bug is like leaving the workshop’s storage room door unlocked: outsiders can walk in without a login and either read the contents or delete messages. If your business uses Langflow publicly or on a reachable network, this can directly expose and destroy your users’ data.

This matters if you run langflow or langflow-base and expose its API endpoints on a network where other people can reach them. The bug does not require login and does not require someone to “click” anything for the attack to work. Because it depends on network reachability, you should treat it as a risk if your Langflow server is reachable from the internet or otherwise reachable by untrusted parties.

This is RED because attackers don’t need an account to access and delete sensitive user data/messages—making it straightforward to exploit once endpoints are reachable. The advisory is driving increased attention, and the fix has specific patched versions; you should prioritize upgrading right away.