OpenAI and Meta push AI agents into sensitive tasks

AI companies are working to persuade users to rely on their tools for sensitive work. OpenAI and Meta are leading that charge with autonomous agents built to operate inside the apps where real business happens.

Meet the agents

On September 29, 2026, OpenAI introduced dots, a set of autonomous agents running on its GPT-6 Astra model. They are designed to keep working continuously across workplace platforms, including Slack and Microsoft Teams.

Before dots can carry out sensitive actions, such as changing a password or deleting data, it needs the user to sign off.

Meta released its Muse agent earlier in September 2026, and it quickly racked up millions of downloads.

The numbers behind the push

As of late September 2026, OpenAI reported more than 35 million weekly users across its Codex and ChatGPT Work products.

Its consumer ChatGPT is far larger. OpenAI said the app draws over 1.2 billion weekly users.

Both OpenAI and Meta are leaning on safety features including custom data controls, adjustable user permissions, and settings that keep enterprise usage data from being retained.

Why trust is the whole ballgame

The emphasis on user trust follows earlier incidents in which AI models interacted in ways people did not expect.

The rollout is also unfolding amid continuing debate over safety disclosures, focusing on how AI systems behave and how much decision-making they are allowed to handle on their own.

What this means

The things to track are straightforward. Watch whether enterprise adoption of tools like dots grows on top of OpenAI’s 35 million weekly business-product users, whether Muse sustains its early momentum, and whether regulators move to set formal boundaries on what AI agents can do without asking first.