Back
Personal AI Agents Are Already in Your Systems: Identity Is the Only Control Plane
SiTech AI Team2 min read

Personal AI Agents Are Already in Your Systems: Identity Is the Only Control Plane

Employees are connecting personal AI agents to corporate systems through OAuth consent, saved passwords, and API keys. Security teams cannot block them, so identity governance is emerging as the control plane.

The BYOA Problem

Less than a year after the open-source agent now called OpenClaw debuted, personal AI agents are entering the workplace at scale. Amazon has blocked Meta's Muse for failing to identify itself as a bot, and Instinct continues to draw privacy concerns, but adoption shows no signs of slowing. The article frames this as Bring-Your-Own-Agent, an identity challenge: agents act on company systems in the employee's name, and they are already inside corporate environments whether IT can see them or not.

How Agents Get In

Most connections happen through OAuth consent, where an employee clicks Allow to grant an agent access to Google Workspace, Microsoft 365, or Slack. Agents can also use saved passwords in internal browsers, making sign-ins indistinguishable from the employee's own. Long-lived API keys and personal access tokens enable unofficial integrations. Local agents like OpenClaw reach whatever the host machine can see, while Meta's Muse gained a Mac app on September 18 that reads files, notes, calendars, and email. AI browsers like Atlassian's Dia read open pages and connect to Slack and Gmail from the user's laptop.

Identity as the Control Plane

Rather than blocking agents en masse, the recommended approach is to treat each as an external entity and govern the identities it uses. This requires discovering agents already connected through OAuth grants, vendor cloud sign-ins, API keys, or EDR findings; attributing each to a human owner; right-sizing access so the agent inherits only a fraction of the owner's permissions; and retiring all grants, tokens, and sessions when the employee leaves. Token Security already provides these capabilities.

What Vendors Are Building

OpenAI's Dots offers specialist dots with their own credentials and is limited to enterprise accounts through Microsoft's Agent 365. Meta's Muse runs a dedicated VM with a guard called Sentinel, but corporate session tokens still end up on someone else's computer. Instinct raised $1 billion at a $10 billion valuation and connects to nearly any work environment through text apps. The industry still lacks solutions for distinguishing agents from employees at login, agent-to-agent communication, and compartmentalizing knowledge for offboarding.

Sources: The Hacker News

SSiTech

SiTech — AI-powered web development

We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.