v0.74.1
July 26, 2026
Protection against connection hijacking now covers calls to AI providers too. The previous version bound outbound connections to already-verified addresses for webhooks, n8n and email; the path to AI providers was still on the old transport. It now follows the same rule.
-
Security
Calls to AI providers now use the protected transport
Time can pass between the moment an address is checked and the moment the connection opens, and a hostile system controlling its own domain can use that window to send the call somewhere else — typically towards the internal network. It's the same window closed in the previous version for other outbound channels; on the AI path it was still open. Where the network policy is strict, the connection is now bound to the verified address. If you run a model on your own network, such as Ollama, nothing changes: those installations keep working as before, because always binding would have broken them.