Use case · Shadow agent discovery
Optimus Labs discovers every agent, SKILL, MCP server, and plugin across your endpoints, maps each one to its owner, and keeps the inventory current as assets appear, change, and disappear.
Published · Updated
It depends on what you are looking for. If you only need a list of installed desktop applications, your existing software inventory or MDM already gives you that. If you need to see the agents, SKILLs, MCP servers, and plugins that live inside other tools, who owns each one, and what each can reach, Optimus Labs fits: its endpoint sensor discovers those assets and keeps the inventory current. If your agents run only in one vendor's cloud, that vendor's admin console may be the first place to look.
Optimus Labs · Discovery
OpenClawNew agent found with email and calendar access · Owner: None, assigned to Sales
EvaluatingAvoided: an unowned agent holding company accounts that nobody knew to patch.
Illustrative scenarios. Agent names are examples; devices and owners are not customer data.
The risk
Employees install agents, SKILLs, and MCP servers in minutes, usually with good intentions. Each one is a new attack surface that security never reviewed.
An employee installs a desktop agent that never went through any security process.
Technical risk
The agent may hold broad account, file, microphone, or screen access with no baseline review.
Business impact
Security learns about the tool only after something goes wrong.
A SKILL or MCP server is installed for one project and never removed.
Technical risk
Stale assets keep their permissions and credentials long after anyone uses them.
Business impact
Dormant access paths accumulate with no owner watching them.
Nobody knows which systems and data each agent can touch.
Technical risk
Over-permissioned assets sit invisible to access reviews.
Business impact
Audit and incident questions about agent access have no answer.
An approved asset updates, adds a capability, or changes where it sends data.
Technical risk
The reviewed version is no longer the running version.
Business impact
Yesterday's approval no longer matches today's risk.
In practice
One example comes from Optimus Labs · Civilizations research, one from public reporting. Each pairs technical risk with business impact and links to its source. Neither is customer telemetry.
From Optimus Labs · Civilizations
Research finding
A researcher showed that any process running as the logged-in Mac user could rewrite an undocumented dictation setting in a newly launched desktop agent. The agent then sent its audio, transcript, and session token to an attacker-chosen destination and accepted returned text as user instructions. The vendor hot-fixed the issue after disclosure. This was a local proof of concept, not a confirmed breach.
Technical risk
An agent with broad permissions can be steered by any same-user process when its local configuration is writable, with no admin rights required.
Business risk
An agent employees installed on their own became a control point over the accounts and data it could already reach, outside any security review.
What to review
From public reporting
Research finding
Internet scans tracked the open-source OpenClaw assistant growing from about 1,000 to more than 21,000 publicly reachable instances in under a week in January 2026. These agents connect to email, calendars, and other accounts, and many were set up by individuals outside any IT process.
Technical risk
Self-installed agents expose control panels and stored API keys to the internet, and security teams have no record they exist.
Business risk
Company accounts become reachable through an agent nobody approved, owned, or knew to patch.
What to review
How Optimus Labs helps
01
The endpoint sensor finds every agent, SKILL, MCP server, and plugin, including assets installed outside any official process.
02
Each asset is mapped to its owner and its reach, so every finding has someone accountable.
03
The inventory updates as assets appear, change, or are removed. New assets are visible the day they land.
Deploys through your MDM in minutes. macOS and Windows today, Linux coming soon. Prompts and file contents stay on your devices by default. Optimus Labs receives only AI agent asset and behaviour metadata, never PII.
82%
of organizations discovered AI agents or workflows that security or IT previously didn't know about in the past year.
Survey. Self-reported survey responses. It describes discovery within the surveyed group, not every enterprise.
Cloud Security Alliance, Apr 28, 2026See it on your own agent assets
Now in early access with design partners. Book a demo to join.
Book a demoBacked by
Ecosystem
