Rapid7 · an outside read from public information · November 2026
One testable bet: extend Kenzo to correlate an attack across a customer’s AI agents, with a person approving containment.
Three challenges (from public reporting)
- Fragmented data. Each product, and now each agent vendor, logs only its own slice.
- Non-core products. About 15% of ARR, down about 19% a year, with end of sale announced.
- Short window. Autonomous SOC v1.0 is planned around RSAC 2027.
Five possible next moves
- Extend Kenzo to secure AI agents: feed coding agents, assistants and in-house agents into the same data mesh as identity, cloud, SaaS and egress data, vendor-neutral.
- One approval, every tool: human approval and an audit log that can contain an agent chain across vendors.
- Shadow mode first: agents recommend beside analysts, then compare.
- A measured pilot: one agent workflow, scored on time to containment, false-containment rate and correlation accuracy.
- AI cost per investigation, tracked next to quality, with pricing guardrails.
Public MCP/agent-integration releases
| Vendor | Release | Source |
|---|---|---|
| Rapid7 | Open-source Bulk Export MCP server: vulnerability and exposure data, read-only | [7] read 15 Nov 2026 |
| CrowdStrike | Falcon MCP server; moving to a hosted version | [10] read 15 Nov 2026 |
| Palo Alto Networks | Cortex MCP server for XSIAM cases, issues and assets | [11] read 15 Nov 2026 |
| Microsoft | Sentinel MCP server, generally available | [12] read 15 Nov 2026 |
| Tenable | Hosted Hexa AI MCP server over exposure data | [13] read 15 Nov 2026 |
Coding agents already ship permission modes, sandboxing, MCP allowlists, audit logs and telemetry export. Each covers its own sessions and exports to the customer's own collector or SIEM, where the cross-vendor picture has to be assembled.
Reviewed with GPT-6.1 Sol Pro, Claude Fable 5.2, Gemini 4 Argon · public info only · synthetic data.
This link deactivates 24 hours after sharing.