Skip to content
Rush Commerce
AI & Automation3 min read

Opsin raises $21M for AI agent security: 93% had full access

Opsin raised $21M for AI agent security after finding 93% of misaligned agent tools had full system access and 1 in 200 needed approval. Scope yours now.

Opsin raised $21 million to watch what AI agents do inside companies, and its own data is the story. In the funding announcement, the AI agent security startup says that 93% of the misaligned tools connected to agents in its customers' environments had "full, unrestricted access" to the systems they touched. Only 1 in 200 required human approval before taking action.

What actually happened

Opsin announced the round on October 15. SYN Ventures and Race Capital led it. The facts from the release:

  • The product. Opsin finds AI agents running across an organization and links each one's purpose, permissions, data access, tools and activity. The goal is to flag access that is too broad and behavior that drifts from the agent's job.
  • Where it works. Agents built on Microsoft Copilot, Claude, ChatGPT Enterprise, Gemini and other approved tools.
  • Customers. Nearly 20 enterprises, including Culligan, Barry-Wehmiller, Wellstar Health System and UiPath.
  • Team. CEO James Pham and co-founder Oz Wasserman both came from Abnormal Security. Jeremy Mailen is the third co-founder.

One caveat. The 93% figure comes from a vendor that sells the fix, and it is measured on "misaligned" tools, not all agent tools. Read it as a direction, not a census.

Why AI agent security matters for your business

You do not need a $21M platform to act on this. The failure Opsin describes is simple: someone connects an agent to Google Drive, the CRM or the store admin with the easiest permission, which is usually "everything." Then nobody looks again.

A small business has the same problem at smaller scale. Your support agent can probably read every customer record. Your bookkeeping automation can probably edit, not only read. That is fine until a prompt injection, a bad instruction or a model swap changes what the agent decides to do.

Three things to do this month:

  • List every agent and every connection. Name the owner, the tool and the scope of each one.
  • Cut each scope to the job. Read-only where possible. One folder, not the whole drive.
  • Put a human gate on the actions that cost money. Refunds, payments, deletes and bulk emails get an approval step. Approval on everything gets ignored, so put it where the damage is.

Key takeaways

  • Opsin raised $21M, led by SYN Ventures and Race Capital, on October 15
  • Its data says 93% of misaligned agent tools had full, unrestricted system access
  • Only 1 in 200 required human approval before taking action
  • The figures are vendor data, but the pattern is common: agents get the broadest permission by default
  • Inventory your agents, cut scopes to the job and gate the actions that cost money

An agent should have the keys to its job, not to the building. We build AI agents with scoped credentials, approval steps on the money moves and an audit log you can read. See how we build them, or ask us to review your setup.

Sources: Opsin via Newswire, StartupResearcher.

  • #ai-agents
  • #agent-security
  • #ai-governance
  • #permissions
  • #funding
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.