An AI Agent Builds an Unauthenticated SharePoint RCE Chain; the Published PoC Hits Honeypots Within Days (CVE-2026-55040, CVE-2026-63520)

Microsoft SharePoint Server 2016, SharePoint Server 2019, and SharePoint Server Subscription Edition; SharePoint Online is unaffected

Rapid7 Labs disclosed a two-vulnerability chain reaching unauthenticated remote code execution, published on August 11, 2026 with proof-of-concept code. CVE-2026-55040 (CVSS 9.1) is a set of failures in SharePoint's JWT token validation pipeline that let a remote unauthenticated attacker assume a chosen user's identity, including an administrator, given only that user's Active Directory SID or user principal name. CVE-2026-63520 (CVSS 8.1) is unsafe .NET type instantiation in Business Connectivity Services, giving code execution as the Windows service account. Microsoft patched the first in July 2026 and the second in August. The research existed to test whether an AI agent could find and weaponize a real exploit chain in a mature codebase: across two sprints, in January and March 2026, the January attempt produced nothing usable and March produced the chain, over 24 active days spanning 96 sessions, 256 prompts, and roughly 80,000 agentic tool calls. Rapid7 was direct that full automation did not work, and that human researchers had to steer continuously because the model too often produced findings that were questionable or inaccurate. Within days of the PoC going public, threat intelligence firm Defused observed attackers using it against SharePoint honeypots; Microsoft had not confirmed in-the-wild exploitation while flagging the flaw as a likely target

Conventional authentication and deserialization defects in a widely deployed enterprise product, found faster than human-only review would have found them. The security-relevant fact is the shape of the effort: 80,000 tool calls with a human in the loop the whole way, and a public PoC that attackers picked up within days of publication