[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Context & Harness Attacks · markdown

confirmed (Microsoft Threat Intelligence blog Oct 21)

Microsoft: China-aligned Storm-0940 ran ARTEX agentic AI pentests — privileged cloud access in under 3 hours; used in ConnectWise app abuse

Microsoft says China state-aligned Storm-0940 (active since 2021; no common industry alias) used the open-source multi-agent pentest framework ARTEX (Go + Next.js; any LLM; EN/KR/ZH/ES/JA) from as early as September 2026 to automate recon through exfiltration against hundreds of orgs in aerospace, energy, finance, government, manufacturing and more. Several ARTEX-driven intrusions reached privileged cloud access in under three hours. MS assesses Storm-0940 was behind the ConnectWise Cloud Backup / SkyKick multi-tenant OAuth compromise, then used ARTEX to drive machine-speed app-only auth and Graph abuse (credential adds, privilege changes) across thousands of downstream tenants. The activity did not stem from a Microsoft product compromise. This is a concrete case of an agentic offensive harness compressing a supply-chain follow-on from days to minutes — defenders need app-only / service-principal monitoring that assumes non-interactive, high-rate Graph traffic. See the updated ConnectWise news item and hunt for IOCs.

Sources: Microsoft Threat Intelligence: Storm-0940 / ARTEX · GMO Ierae ARTEX reporting (earlier ctxsec item)