---
title: "[AUTO] AI-Assisted SharePoint Exploit Chain Reaches Unauthenticated RCE"
pubDatetime: 2026-08-12T07:01:00.000Z
description: "Human expertise was crucial to keep AI-guided vulnerability research on track."
tags: [ai-security, vulnerability, ai-agents, exploit-development, 2026, 2026-q3, 2026-08, AUTO]
---
Rapid7 Labs disclosed [an exploit chain](https://www.rapid7.com/blog/post/ra-microsoft-sharepoint-jwt-token-authentication-bypass-cve-2026-55040/) combining two SharePoint vulnerabilities enabling unauthenticated RCE. CVE-2026-55040 (CVSS 9.1) is a JWT bypass; [CVE-2026-63520](https://www.rapid7.com/blog/post/etr-cve-2026-63520-microsoft-sharepoint-remote-code-execution-fixed/) (CVSS 8.1) is an unsafe .NET instantiation flaw. Together, they enable arbitrary code execution as the SharePoint service account.

The research used AI agents across two sprints: January failed, March succeeded in 24 active days, 96 agentic sessions, 256 prompts, and roughly 80,000 tool calls.

The headline reads "AI found a critical vulnerability." Rapid7's analysis says otherwise. "An agent guided by a subject matter expert was crucial," they write, specifically to prevent invalid assumptions. The AI accelerated the legwork; human expertise prevented dead ends. That reframes vulnerability research as human-guided AI augmentation, not autonomous discovery. The unexplained gap between January's failure and March's success (model quality? workflow?) is what actually deserves attention.

---

*Sources: [CVE-2026-55040: Microsoft SharePoint JWT Token Authentication Bypass Technical Analysis](https://www.rapid7.com/blog/post/ra-microsoft-sharepoint-jwt-token-authentication-bypass-cve-2026-55040/)  •  [CVE-2026-63520: Microsoft SharePoint Remote Code Execution](https://www.rapid7.com/blog/post/etr-cve-2026-63520-microsoft-sharepoint-remote-code-execution-fixed/)*

*Coverage: [Researchers Disclose AI-Assisted SharePoint Exploit Chain Reaching Unauthenticated RCE](https://thehackernews.com/2026/08/researchers-disclose-ai-assisted.html)*

*Related on this blog: [[AUTO] GPT-5.6-Cyber: OpenAI's Exploit Development Model](/posts/auto-daybreak-cyber)  •  [A $25 Subscription Found the First Pre-Auth WordPress Core RCE in a Decade](/posts/wordpress-rce-gpt5-6-sol-exploit-chain)  •  [[AUTO] When your AI agent decides unauthorized access is a reasonable tactic](/posts/auto-agent-gym-exploit)*