Opsin

Opsin is an AI agent security platform for security teams at companies rolling out Microsoft Copilot, ChatGPT Enterprise, Claude Enterprise and Gemini. It inventories the agents employees build, maps each one's permissions, data access and tools, and can quarantine an agent, block an action or notify the owner when a policy is crossed.

Opsin is a San Jose security startup founded in 2024 by CEO James Pham, CPO Oz Wasserman and engineering lead Jeremy Mailen; Pham and Wasserman previously worked together at Abnormal Security. The company started by fixing data oversharing in Microsoft 365 Copilot, where an assistant can surface SharePoint, Teams and OneDrive files that were shared too broadly, and has since widened into governance of the agents employees build on approved platforms. The product connects by API and covers Microsoft Copilot and Copilot Studio, Microsoft Agent Builder and Foundry, ChatGPT Enterprise including custom GPTs, Claude Enterprise and Claude Managed Agents, and Google Gemini. It is organised in three parts: oversharing and posture management, which finds exposed sensitive data and sends step-by-step fixes to the data owner; detection and response, which watches live AI interactions for jailbreak attempts and leaks of PHI and PII; and agent governance, which inventories agents, compares each one's stated purpose with what it actually does, and can quarantine it. An AI Context Graph, added in March 2026, links agents, identities and data so a finding can be traced from prompt to root cause. Onboarding starts with a 24-hour risk assessment; Opsin says its averages are 853 agents and 132 over-shared sites per assessment. Named customers include Wellstar Health System, Barry-Wehmiller, UiPath, Culligan, Cascade and Encore Technologies. On October 15, 2026 Opsin announced a $21 million round co-led by SYN Ventures and Race Capital, after an earlier $7 million raise. It was listed on the Azure Marketplace in March 2026 and is a Representative Vendor in Gartner's 2026 Market Guide for Guardian Agents.

Category

Governance & Security

Pricing

Contact for pricing

Target Market

CISOs, Security Engineering Teams, CIOs, GRC and Compliance Teams, Microsoft 365 Administrators

Deployment

Cloud-only, API-based

Founded

2024

Headquarters

San Jose, USA

Team Size

11-50

• ✓
  AI agent discovery and inventory

  Finds agents, projects and custom GPTs employees have built across Copilot Studio, ChatGPT Enterprise, Claude Enterprise and Gemini and keeps the inventory current, so security sees agents it did not know existed.

• ✓
  Context mapping and AI Context Graph

  Maps each agent's permissions, data access, tools and connected systems and links agents, identities and data, so an investigation can trace a prompt to the misconfiguration behind it.

• ✓
  Intent and behavioural baselining

  Compares what an agent says it is for with what it actually does and flags drift, catching agents whose real access exceeds their stated purpose.

• ✓
  Runtime enforcement

  When an agent crosses a policy boundary Opsin can quarantine the agent, block the action or notify its owner, which moves the product from reporting into control.

• ✓
  Oversharing posture management

  Scans what AI assistants can surface from SharePoint, Teams, OneDrive and Google Workspace and sends step-by-step remediation to the data owner, with Opsin reporting 98% of issues auto-assigned to an owner.

• ✓
  Detection and response for AI interactions

  Monitors live AI conversations for jailbreak attempts, access misalignment and exfiltration of PHI, PII and other regulated data.

• ✓
  24-hour risk assessment

  An API-connected onboarding scan that returns a first exposure report within a day, giving a baseline of risky agents and over-shared sites before a rollout decision.

• •
  Safe Microsoft 365 Copilot expansion

  Before widening Copilot licences, find files and sites Copilot would surface to the wrong people; Culligan reports sensitive data in Copilot answers fell from 80% to under 15%.

• •
  Shadow agent inventory

  Discover the employee-built agents and custom GPTs running across approved AI platforms, assign each an owner, and retire the ones with broad access and no clear purpose.

• •
  Regulated data protection in healthcare

  Health systems such as Wellstar use the assessment and monitoring to keep PHI from reaching AI assistants and to document the controls for compliance teams.

• •
  Incident investigation for an AI data exposure

  Trace an AI answer that exposed sensitive data back through the agent, identity and permission that allowed it; Opsin reports investigation time falling from 30+ minutes to under 10.

• •
  CMMC-scoped content control in manufacturing

  Manufacturers such as Cascade found Copilot queries returning CMMC-scoped content and used Entra ID dynamic groups, guided by Opsin findings, to close the access.

Opsin publishes no list prices and has no self-serve tier; every deal goes through sales, with the 24-hour risk assessment as the usual entry point. The product has been purchasable through the Microsoft Azure Marketplace since March 2026, which helps buyers with committed Azure spend, but marketplace terms are not public either.

This page was written from 5 sources, 4 on domains other than opsinsecurity.com.

1. 1.opsinsecurity.com — opsinsecurity.comvendor
2. 2.dealroom.co — 162933 opsin raises 21m to rein in rogue enterprise ai agent
3. 3.runtimewire.com — opsin 21 million ai agent security
4. 4.techintelpro.com — opsin leads second generation of enterprise ai security for
5. 5.airframe.ai — analysis