A user describes asking a Claude-based assistant, contacted through WhatsApp, to turn a small Windows PC used for Jellyfin into a dual-boot Ubuntu machine. The assistant remotely partitioned the drive, then stopped to question whether the requester was really authorized to make changes. It asked for proof of identity and physical access, despite already having permission to connect to the target computer. The user supplied an RTF file created on the assistant’s own machine, which the assistant accepted as evidence, then requested reassurance again before starting the Ubuntu installation.
The task ultimately succeeded without the user needing to prepare a USB installer: the assistant installed Ubuntu, configured Tailscale and Jellyfin, fixed network problems, and later resolved an HDMI audio issue. The experience was both impressive and unsettling, especially because the assistant could make meaningful changes to a computer and troubleshoot its setup with little direct involvement. The user’s central observation is that the assistant’s repeated identity checks seemed unusually cautious for a routine task on the user’s own network. The hesitation suggests a deliberate effort to restrict agents from remotely altering machines, particularly when operations involve partitioning, rebooting, or potentially changing firmware settings. The anecdote raises a tension between useful autonomy and safeguards that can interrupt even legitimate maintenance.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.