Spent an hour last night debugging "the internet is down, but intranet works" on a Mac mini, our in house container for dozens of 24/7 agent loops ping worked. SSH worked. Existing connections worked. Every new TCP connection died instantly with "Can't assign requested address." I chased DNS, the firewall, the proxy, ephemeral ports, kernel buffers. All wrong. at one point all evidence pointed to a potential bug caused by one component in Phi Browser, i created an urgent issue and assigned to one of our devs ... Uptime: 49d 19h 36m macOS keeps its TCP clock (tcp_now) in a uint32 of milliseconds. At 2^32 ms (49d 17h 02m 47s) it wraps, a monotonicity check decides the new time is "in the past", and the TCP clock freezes forever. TIME_WAIT stops expiring, ephemeral ports leak away, new connections die. Old ones live on, which is what makes it so hard to see. We were 2.5 hours past the fuse. TIME_WAIT that keeps growing while nothing is running. Confirmed on macOS 26.2, M4. Only a reboot fixes it. Until Apple does something to fix this shenanigan, only fix is: reboot your 24/7 mac every 49 days