OpenAI Launches Persistent Agents With Permission Controls and Separate Action Review
2026-09-29 ยท Safety & Alignment
OpenAI introduced dots, persistent agents with cloud workspaces and app access, alongside Auto-review, a system OpenAI says checks consequential actions against instructions and permissions before they execute.
OpenAI introduced dots on September 29: persistent agents with cloud workspaces and access to a user's connected applications and, optionally, a personal computer. OpenAI's accompanying safety documentation describes Auto-review, a separate system that checks consequential actions against a user's instructions and permissions before execution, with the controls enforcing those checks sitting outside the environments the agents themselves can modify.
These are OpenAI's own claims about its own architecture. The published design describes intended protections; it does not by itself establish their effectiveness, and no independent evaluator's assessment of dots or Auto-review is cited here.
For SIRDA, this bears directly on the combination the site's own framing tracks: persistent autonomy, access to digital systems, delegated authority, and controls on consequential actions. Auto-review is a separate software review system and should not be conflated with NVIDIA's hardware-based agent monitor, Sentry, covered in a different Recently Active entry.