Personal AI agents have moved from demos to daily use in a matter of weeks. Meta launched Muse, and OpenAI followed with Dots at DevDay. Employees are putting these agents to work on both personal tasks and business workflows.
OpenAI Dots joins a growing roster of workforce, coding and personal agents Aembit secures, including ChatGPT, Claude, Gemini, Microsoft Copilot Studio and Meta Muse. Aembit customers can secure Dots today with the same Aembgmit MCP Identity Gateway they already use, with no new product to deploy.
What Is OpenAI Dots?
OpenAI Dots is an always-on personal AI agent that OpenAI introduced at DevDay on Sept. 29. A Dot works toward goals you give it in the background, using its own cloud computer and browser to carry out multistep tasks over hours or days. Users can interact with Dots through ChatGPT, Slack and Microsoft Teams and connect them to apps and services through plugins.
Why Personal AI Agents Need Identity and Access Controls
Major AI platform vendors are now shipping personal agents, and people use them for both personal and work tasks. The same Dot that plans a weekend trip might also triage a GitHub backlog, summarize a customer thread or pull numbers from an internal system. That overlap between personal and professional use can leave security teams with limited visibility into enterprise access.
Organizations want employees to benefit from these agents, and a ban can drive usage underground. Security teams need control to avoid “shadow agents,” namely the ability to let Dots reach enterprise resources while deciding what each agent can access, under which conditions and with a record of its activity.
Access based solely on an employee’s authorization can make that harder. A personal agent may rely on delegated OAuth permissions and persistent credentials, making its activity difficult to distinguish from the employee’s. Without separate agent controls, security teams may struggle to enforce tighter limits or revoke agent access independently. An always-on agent makes decisions at runtime, so its access controls need to account for more than the permissions granted at setup.
How Dots Connects to the Aembit MCP Identity Gateway
OpenAI Dots and ChatGPT share the same plugin settings for connecting to outside services. If you already secure ChatGPT with Aembit, Dots follows the same path.
Instead of configuring a separate connector and set of credentials for every MCP server a Dot might need, the user adds a single connector that points to the Aembit MCP Identity Gateway. Behind that connection, Aembit manages access to MCP servers for services such as GitHub, Microsoft 365 and Salesforce, along with your own internal tools.
Here is what happens on each request:
- The Dot connects to the Aembit MCP Identity Gateway through its connector and authenticates.
- The Aembit MCP Authorization Server evaluates policy based on blended identity: the agent, the employee it acts for, group membership and conditions such as security posture.
- If policy allows access, Aembit issues a short-lived, scoped credential for the target MCP server at runtime. The Dot does not need to hold a long-lived credential for that server.
- MCP Tool Access Control filters which tools the Dot can discover through
tools/listand checks eachtools/callbefore it reaches the server. - Access decisions and tool calls through the gateway are logged centrally.
What Aembit Customers Get
- A distinct identity for every Dot. Each agent has its own identity, linked to the employee it works for, so you can distinguish agent activity from human activity.
- Just-in-time access. Credentials are issued at runtime, scoped to the request and expire quickly, so the agent does not need to retain static credentials for connected MCP servers.
- Least privilege down to the tool. MCP Tool Access Control lets you allow a Dot to read from a server without letting it delete data, make payments or take administrative actions. Rules support exact tool names or patterns, and enforcement fails closed.
- One connection, many MCP servers. A single Dots connector to the Aembit MCP Identity Gateway replaces separate connectors and credentials for each server.
- A centralized audit trail. Policy decisions and tool calls through the gateway are logged in one place for investigations and compliance reviews.
- A kill switch that leaves the employee working. Security teams can revoke a specific agent’s access through Aembit without locking the employee out of their own accounts.
- Consistent policy across agents. The same policy framework covers requests from Dots, ChatGPT, Claude, Copilot Studio and Muse.
Access Controls Across Your AI Agents
Dots is part of a growing list of agents Aembit secures: workforce agents like Microsoft Copilot Studio, coding agents and assistants built on Claude and the OpenAI API, general assistants like ChatGPT and Gemini, personal agents like Meta Muse and OpenAI Dots, and custom agents your teams build in-house.
These agents connect to the same identity control plane, so your policies and audit trail stay centralized as adoption grows.
Get Started Securing OpenAI Dots
Existing Aembit customers can secure Dots today at no additional cost. If you already connect ChatGPT to the Aembit MCP Identity Gateway, use the same gateway connection and extend your access policies to cover Dots. Your Aembit team can help with setup.
New to Aembit? Start with the Agentic AI Deployment Checklist or book time with an Aembit engineer to see how Aembit secures Dots access.
