AI agent security: prompt injection, permissions, sandboxes, supply chain
Essence
An agent acts on your behalf and with your privileges: it reads files, runs commands, goes to the network, pushes to git. So the question isn't "is the model smart" but "what happens if the model makes a mistake or gets tricked". Safe work rests on three principles: least privilege, isolation and human review of anything that leaves your machine.
The rest of this answer, with code and interview follow-ups, is in the DevPrep app.
Read the full answer in DevPrep