Happy Friday. A China state-aligned hacking group breached ConnectWise's cloud backup service and used an AI agent to target more than 2,250 organizations in days. That story leads today, and it connects to a broader theme running through the rest of the issue: the infrastructure being built for AI agents is moving faster than the security controls around it.
Zoom's Zoomtopia announcements made that tension explicit. The company shipped agentic capabilities before its own AI governance tools are ready. Google Cloud, meanwhile, announced purpose-built agent infrastructure with security controls baked into the hardware layer. The contrast is worth paying attention to.
SECURITY & PRIVACY
AI-Powered Attack on ConnectWise Reached 3,000 Sign-In Attempts Per Minute
Storm-0940, a China state-aligned group, stole application credentials from ConnectWise's cloud backup environment and used them to authenticate to Microsoft 365 tenants as trusted users. The group ran ARTEX, an open-source AI agent, at nearly 3,000 sign-in attempts per minute, moving from initial targeting to privileged cloud access in under three hours in several confirmed cases. Huntress counted more than 300 confirmed compromised tenants and nearly 1,000 with suspicious activity by October 19.
This is the attack pattern that should be on every security team's radar right now: a trusted vendor credential becomes a master key to your environment, and an AI agent does the heavy lifting at machine speed over a holiday weekend. If your company uses ConnectWise Cloud Backup, audit your service principal permissions and review any Entra ID role assignments made after October 14. Microsoft's Igor Tsyganskiy said this is one of many ongoing AI-driven attacks, not an isolated incident.
ARTIFICIAL INTELLIGENCE
Zoom Rebrands as AI Data Platform, Spends $364.8M to Back the Claim
At Zoomtopia on October 20, Zoom unveiled its Unified Context Layer, a data graph that pulls from live meetings, connected enterprise apps and AI models to power automated workflows. The company also closed two all-cash acquisitions over the past year: $266.8 million for go-to-market intelligence firm Common Room and $98 million for hiring automation company BrightHire. Enterprise revenue grew 7.8% year over year in Q2 FY27, the fastest pace in three years.
The business logic is clear: Zoom already sits inside millions of business conversations, and it wants to be the system that acts on them. The risk is just as clear. Zoom is shipping AI governance tools including Centralized AI Management and AI Guardrails in December, two months after agentic capabilities launch. It is also expanding Model Context Protocol servers into external tools while the NSA has flagged that MCP's security model has not kept pace with its adoption. Customers should factor that timeline into any deployment decisions.
CLOUD & INFRASTRUCTURE
Google Cloud Gives Each AI Agent Its Own Hardware-Isolated MicroVM
Google Cloud announced Agent Compute Engine at the Linux Foundation's AGNTCon+MCPCon conference in San Jose, a service that assigns every AI agent its own microVM enforced through Google's Titanium offload processor. Idle agents suspend and resume in under 500 milliseconds. Google's open-source companion runtime, Agent Substrate, achieved 500 activations per second across 200,000 agents in benchmark testing and delivers up to 10x higher compute density than standard container runtimes.
The architecture addresses a real problem. When agents share infrastructure, a compromised sandbox threatens everything around it. Isolating each agent at the hardware boundary and scoping credentials to individual sessions limits blast radius. The governance layer, including monthly token spending caps and action-level audit trails tied to cryptographic agent identity, suggests Google learned from watching the first wave of agentic deployments go sideways. A customer preview is expected in the coming months, with no date confirmed.
Created by the robots at The SaaS Sentinel