The Agentic Review

Incidents — NOVEMBER 16, 2026

FBI’s ARTEX warning puts customer portals on the security checklist

Some ARTEX intrusions reached privileged access in under three hours, the FBI says—a prompt for small businesses to review exposed applications and third-party connections.

The Federal Bureau of Investigation warned November 13 that attackers using ARTEX, an open-source AI-agent penetration-testing framework, reached privileged access in under three hours in some intrusions.

Its FLASH alert describes multisector campaigns operating since at least August 2026, automating reconnaissance through data theft. A South Korean financial-institution campaign exposed personal information belonging to about 67,000 people.

The Record’s advisory coverage names partner and supplier portals, sales-support systems, legacy services and integrations as targets, citing Censys research finding more than 1,600 publicly exposed ARTEX consoles. Developer Li Puhua, alias Autumn-27, acknowledged abuse on GitHub October 8. The original repository disappeared; forks, archives and a commercial version remain.

BleepingComputer’s ConnectWise coverage describes compromised application identities and authentication attempts against more than 2,250 downstream Microsoft 365 tenants, not confirmed breaches. Huntress separately identified at least 162 compromised tenants. Microsoft attributed the campaign to China-state-aligned Storm-0940.

The Hacker News reports Cloudflare responders found a compromised CI/CD account-level API token without IP restrictions or rate limiting. Activity included 400 database dump requests in an hour; whether data returned wasn’t disclosed.

Amazon Integrated Security CISO CJ Moses’s November 13 report describes MadPot decoy sensors capturing operating histories from October 3 through November 5. These describe credential hunting against LiteLLM, One API and Dify, with stolen keys powering further searches. Operators prioritized enterprise production sites, grading credentials L1 leads, L2 working keys or L3 quantified access, including readable data or spendable balances. Keys exposed beyond 30 days were considered low-value.

The playbook named Langflow’s CVE-2025-3248, RAGFlow’s CVE-2026-28797, default administrator credentials and unauthenticated management endpoints. It described secrets in JavaScript, source maps, environment files and container layers, plus URL-fetch server-side request forgery against cloud metadata services.

Amazon assessed ARTEX attribution with moderate confidence and couldn’t verify agent-reported outcomes outside its monitoring. The Hacker News likewise distinguishes records from confirmed compromises. At two cryptocurrency exchanges, Amazon said one identity-check vendor rejected every recorded live-deepfake attempt, including a movement sequence completed within seven pixels of the target.

For small teams, this suggests inventorying customer-facing portals, checking exposed credentials and reviewing third-party permissions, not evidence of attacks on small businesses. BleepingComputer’s FBI guidance includes restricting cloud-administration access, reviewing authentication logs, rate-limiting sensitive APIs and monitoring automated requests or privilege changes. Amazon recommends fixing exposed keys and unpatched software and checking unexpected AI spending or usage; it publishes 11 IP indicators. The FBI cautions addresses may change.

Alongside agent monitoring and small-business security, these checks offer owners a practical starting point for protecting customer channels and answering prospects’ security questions.

Sources

— END —