AI Ops: sandboxed, audited and EU-hosted agents

In AI Ops in a Controlled Environment I argued that human-safe is not automation-safe: an agent using an engineer's access works at machine scale, thins the human judgment that made that access accept