Any chance that TestFlight is in this mix?
We tried to reproduce it very hard by performing updates with TestFlight but have had no luck so far. Our only reports are from regular users who are not using TestFlight.
Amazingly, I just had a similar occurrence on my Mac this morning. We use a third-party app in a work-from-home setup, and just this morning, it went crazy and was flickering "connecting->disconnected" very quickly. I see that the state was changed every 100 milliseconds in the system logs. The NetworkExtension complains:
2024-02-21 10:36:31.676860+0100 0x1a427 Default 0x0 518 0 nesessionmanager: [com.apple.networkextension:] Found 0 (0 active) registrations for <tunnel provider> (com.apple.networkextension.packet-tunnel)
2024-02-21 10:36:31.677349+0100 0x1a427 Default 0x0 518 0 nesessionmanager: [com.apple.networkextension:] NESMVPNSession[Primary Tunnel:<provider>:496B2DE4-FD7D-482F-9607-4B3021283266:(null)]: Leaving state NESMVPNSessionStatePreparingNetwork
2024-02-21 10:36:31.677522+0100 0x1a427 Default 0x0 518 0 nesessionmanager: [com.apple.networkextension:] NESMVPNSession[Primary Tunnel:<provider>:496B2DE4-FD7D-482F-9607-4B3021283266:(null)]: Entering state NESMVPNSessionStateStarting, timeout 300 seconds
2024-02-21 10:36:31.678820+0100 0x1a429 Default 0x0 518 0 nesessionmanager: [com.apple.networkextension:] NESMVPNSession[Primary Tunnel:<provider>:496B2DE4-FD7D-482F-9607-4B3021283266:(null)] in state NESMVPNSessionStateStarting: plugin NEVPNTunnelPlugin(<plugin>[inactive]) started with PID 0 error Error Domain=NEAgentErrorDomain Code=2 "(null)"
2024-02-21 10:36:31.678897+0100 0x19c0e Default 0x0 15325 0 VPN: (NetworkExtension) [com.apple.networkextension:] Last disconnect error for <3rd party app> changed from "none" to "The VPN app used by the VPN configuration is not installed"
I obfuscated the name of the app, since I'm not sure I should share that publicly.
Does the error The VPN app used by the VPN configuration is not installed ring any bell? I tried to Google it quickly, but no luck so far.
I'm not sure if it's related to the issues we observe, but from the user perspective, the issue looks exactly the same.