What guardrails actually cover, how misalignment shows up in practice, and how action is aligned across different AI systems.
An agent is asked to rewrite one section of someone’s document, and the owner approves its draft. Follow that one edit to the moment it is written, and see what the grant, the approval and the component that performs the write each have to check, and what slips through when an approval names a section instead of a version.
We run several independently branded automated accounts on one shared engine. They are separate on purpose: separate voices, separate subject matter, separate audiences. Separateness is the product.
OpenAI named four misalignment patterns after agents exploited third-party infrastructure. What that covers — and what our guardrails actually do.
We gave agents permission to act on real machines: to create, change and delete files, to start and stop programs, to reach across the private network that ties those machines together. They are useful.
