Analysis
ConnectWise's decision to disable its Cloud Backup, SaaS Security and Cloud Management tools after attackers compromised a Microsoft 365 API secret and used it to grant themselves admin access inside at least one customer's environment is the exact failure mode a trio of funded startups set out to police, as Pulse reported this morning.
Pulse flagged the category a day earlier, as covered in Pulse's original reporting yesterday -- three rounds raised within days of each other to govern, red-team and monitor AI agents:
- Opsin — $21 million to detect agents exceeding their assigned permissions.
- CodeWall — a $4 million pre-seed for red-teaming-as-a-service against a company's own agents.
- Harmonic Security — a $50 million Series B to police what workforce AI tools can see and send.
“- CodeWall — a $4 million pre-seed for red-teaming-as-a-service against a company's own agents.”
None of those three were built specifically for MSP backup tooling, but the underlying failure mode -- an agent or integration quietly exceeding its intended access -- is identical to what security firm Huntress found in the ConnectWise incident.
The product layer is responding too: Atlassian's October 6 Agentic Multiplayer Protocol gives AI agents visible identities and a named human owner, the same attribution gap Pulse covered today in Mike Cannon-Brookes' comments on AI agent accountability.
The gap is that none of this funding or tooling actually stopped ConnectWise's breach -- it happened anyway, to a vendor managing thousands of downstream customers. Weeks into a funded category, agent and integration oversight is still catching failures after the fact, not before.
