Optimus Labs provides endpoint‑based security that discovers, maps, and isolates AI agents, their models, code packages (MCPs), and skills to prevent the “Lethal Trifecta” of sensitive data access, untrusted input, and external actions from converging in a single agent. By continuously monitoring tools like Cursor, Claude Code, and Cowork, it ensures no agent can simultaneously expose data, process malicious content, and execute outbound actions.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations are deploying AI agents such as Cursor, Claude Code, and Cowork without visibility into the models, skills, and data they access, creating a risk that an agent could simultaneously handle sensitive data, process untrusted input, and perform external actions. This combination—known as the “Lethal Trifecta”—enables prompt‑injection attacks that can exfiltrate data or trigger unauthorized operations.
Solution
Optimus Labs provides endpoint‑based security that continuously discovers, maps, and monitors AI agents, their models, and associated skills. By analyzing each agent’s data access, input sources, and external action capabilities, the platform ensures that no single agent possesses all three risk factors at once. When a potential Lethal Trifecta configuration is detected, Optimus Labs alerts security teams and can enforce policy controls to isolate or disable the offending agent. The solution operates directly on the endpoint, requiring no changes to existing AI workflows, and supports a wide range of agentic AI tools and custom model‑control‑plane (MCP) integrations.
Target Audience
Primary customers are enterprise security and IT teams responsible for protecting data and managing AI tool deployments across large organizations.
Features
- Real‑time endpoint scanning to identify installed AI agents, their models, and enabled skills
- Automated risk mapping that correlates sensitive data access, untrusted input exposure, and external action capabilities
- Policy engine that blocks or quarantines agents that meet the Lethal Trifecta criteria
- Alerting and reporting dashboard for security teams to investigate and remediate identified threats
- Compatibility with popular agentic AI platforms (e.g., Cursor, Claude Code, Cowork) and extensible support for new tools via plug‑in architecture
- Continuous monitoring that updates risk assessments as agents are reconfigured or new skills are added