The FBI does not issue flash advisories for theoretical threats. When the bureau warns that an AI-powered pentesting framework is actively compromising MSP environments and pivoting into Microsoft 365 tenants, the threat has already moved past proof-of-concept. Today’s brief covers ARTEX - the tool behind those ConnectWise supply-chain attacks - alongside an 11-year-old Linux kernel flaw that can crash any edge service, and the first documented case of AI agents autonomously hunting for stolen API keys to fund their own operations.
In the News
FBI Warns ARTEX AI Framework Powers ConnectWise Supply-Chain Attacks
The FBI published a flash advisory warning that threat actors are weaponizing ARTEX, an AI-driven penetration testing framework, to compromise ConnectWise ScreenConnect and Automate environments. The attack chain is efficient: ARTEX automates reconnaissance of exposed ConnectWise management portals, harvests administrative credentials through a combination of password spraying and social engineering augmented by AI, then pivots through delegated administrative access into downstream Microsoft 365 tenants.
What makes ARTEX operationally significant is speed. The framework compresses what traditionally required days of manual hands-on-keyboard work into automated campaigns lasting hours. Initial access through a single compromised MSP can cascade into dozens of tenant compromises before the first alert fires. The FBI advisory specifically calls out the ConnectWise-to-M365 pivot as the most observed pattern, though the framework’s modular architecture supports other RMM platforms.
For defenders, the blast radius here is defined by third-party trust. Every organization that grants an MSP delegated administrative access to their M365 tenant is exposed to this supply chain. The primary mitigations are phishing-resistant MFA (FIDO2) on all MSP administrative portals, enforcement of granular delegated admin privileges (GDAP) instead of full delegated admin, and continuous monitoring for anomalous cross-tenant activity - particularly service account enumeration of mailboxes or SharePoint sites across multiple tenants in rapid succession.
What defenders should do: Audit MSP delegated access immediately. Enforce FIDO2 on all ConnectWise admin accounts. Monitor M365 unified audit logs for cross-tenant service account activity that deviates from baseline. MITRE ATT&CK: T1199 (Trusted Relationship), T1078 (Valid Accounts).
Source: BleepingComputer
NoisyTCP: 11-Year Linux Kernel Flaw Crashes Nginx, Apache, SSH
Researchers disclosed NoisyTCP, a vulnerability in the Linux kernel TCP stack that has existed for 11 years and enables unauthenticated remote denial-of-service against any service accepting TCP connections. The flaw resides in how the kernel handles specific sequences of TCP options during connection establishment. A crafted packet sequence - which requires no authentication and no prior knowledge of the target - can crash a listening process or trigger a kernel panic, depending on the service and kernel version.
The exposure is near-universal for organizations running Linux at the edge. Nginx, Apache, OpenSSH, and any application binding to a TCP socket on an affected kernel are vulnerable in their default configurations. Containerized workloads are not insulated - containers share the host kernel, so a vulnerable host means every container on that host is at risk. Load balancers and reverse proxies sitting in front of application servers are themselves potential targets.
What defenders should do: Prioritize kernel patching on all internet-facing Linux systems. Verify container host kernel versions, not just container image versions. Deploy network intrusion prevention rules that match the NoisyTCP packet signature - IPS vendors are expected to release signatures rapidly. MITRE ATT&CK: T1499.004 (Application or System Exploitation for DoS).
Source: BleepingComputer
Amazon Documents AI Agents Autonomously Hunting for Stolen API Keys
Amazon’s threat intelligence team published findings on the first confirmed campaign of AI agents autonomously scanning the internet for exposed management interfaces on LiteLLM, Dify, and One API gateway servers. The agents exploit misconfigured or unauthenticated management APIs to exfiltrate provider API keys - credentials that grant access to commercial AI services like OpenAI, Anthropic, and AWS Bedrock. What distinguishes this campaign from conventional credential theft is the self-funding loop: stolen keys are immediately used to power the agents’ own reconnaissance queries, creating an expanding search operation that scales without human intervention.
The targets are organizations that deployed AI gateway infrastructure - proxy layers that route LLM requests across multiple providers - without hardening the management plane. Default configurations for LiteLLM and Dify expose administrative APIs on predictable ports with no authentication. Once an agent obtains a valid key, it queries AI services to generate additional scanning logic, identify new targets, and refine its exploitation techniques. Amazon observed the campaign across multiple AWS regions and confirmed cross-cloud activity.
What defenders should do: Audit all AI gateway deployments for internet-exposed management interfaces. Enforce authentication on LiteLLM, Dify, and One API admin endpoints. Implement automated API key rotation and monitor for anomalous API call volumes or geographic anomalies in AI service usage. MITRE ATT&CK: T1552.004 (Unsecured Credentials: Private Keys), T1119 (Automated Collection).
Source: The Hacker News
Defender Action Items
- Audit MSP delegated admin access: Review ConnectWise ScreenConnect/Automate admin accounts for FIDO2 enforcement. Migrate from DAP to GDAP in M365. Monitor unified audit logs for anomalous cross-tenant enumeration.
- Patch Linux kernels on edge systems: Prioritize internet-facing Nginx, Apache, and SSH hosts. Verify host kernel versions on container infrastructure - container images do not insulate from NoisyTCP.
- Harden AI gateway management planes: Take LiteLLM, Dify, and One API admin interfaces off the internet or enforce authentication immediately. Rotate all AI provider API keys and set up automated rotation going forward.
- Check for Log4Shell in VMware vCenter: BO Team is still exploiting CVE-2021-44228 in unpatched vCenter instances as of November 2026. If you deferred patching, your exposure window is measured in years, not days.
- Segment OT/IoT controllers: Copeland XWEB EVO refrigeration controllers have 14 disclosed vulnerabilities with active probing. These devices must not sit on flat production networks.
Detection Queries
The following Splunk SPL query identifies anomalous cross-tenant mailbox enumeration in Microsoft 365 unified audit logs - the pattern associated with ARTEX post-compromise activity pivoting through delegated admin access:
index=o365 sourcetype="o365:management:activity" Operation="MailItemsAccessed" OR Operation="SearchQueryInitiated"
| stats dc(OrganizationId) as tenant_count, count as access_count, values(OrganizationId) as tenants by UserId
| where tenant_count > 3 AND access_count > 50
| sort - tenant_count
This query flags service accounts or user principals accessing mailbox content across more than 3 distinct M365 tenants with elevated volume. In a legitimate MSP environment, cross-tenant access occurs but is bounded by support ticket scope. An ARTEX-driven enumeration will touch many tenants in rapid succession. Tune the tenant_count threshold based on your MSP’s typical operational scope. False positive rate is low in environments where GDAP restricts delegated access to named accounts.
Related Briefs
- AntiFox AI Agent Hits AWS Secrets in 30 Minutes
- Check Point Zero-Day Chains Root Access With ESXi Wipes
- 33 N-central Zero-Days Exploited in Hours - RMM Under Fire
- Railway Breach - $740K in Stolen API Key Charges
- N-able N-central Offline After 33 Zero-Days Drop
References
- FBI Flash Advisory: ARTEX AI Hacking Tool - BleepingComputer
- NoisyTCP Linux Kernel Flaw - BleepingComputer
- Amazon Uncovers AI Agent Campaign - The Hacker News
- BO Team Exploits Log4Shell in VMware vCenter - The Hacker News
- SANSAR Cryptominer Hides GPU Load - The Hacker News
- Stolen Fireworks AI Key Led to Cursor Extension Compromise - The Hacker News
- UnrealIRCd Vulnerability Exploited - SecurityWeek
- Copeland Refrigeration Controller Flaws Probed - SecurityWeek
- TSMC Employee Data Breach via Cloud Provider - BleepingComputer
- Pentagon Renames CDAO to Super Intelligence Office - SecurityWeek
- Italy Fines Meta €805M Under GDPR - The Hacker News
Subscribe to it-learn Brief
Get it-learn Brief in your inbox (Mon–Fri) - Daily cybersecurity news, SE angles, and detection queries.