Daily Cyber Threat Brief — October 19, 2026: AI Agents Break Into 100+ Company Sites With Almost No Human Help

Over the past couple of weeks, NorForge principal threat researcher Snorre Fagerland watched a Chinese-speaking hacker break into websites belonging to more than 100 companies — including Deere, SharkNinja, and Cummins — while doing almost none of the work himself. The operator tasked AI agents with spending hours autonomously probing and breaking into targeted sites. Fagerland found the operation by scanning the web for inadvertently exposed instances of Cairn, an open-source AI problem-solving platform popular with hackers; beginning around October 2 he found about two dozen instances being used for malicious activity, which let him quietly observe the criminals' AI tooling in action. Reuters independently reviewed the hacker's Cairn activity to confirm the victim identities. Fagerland's assessment: real-world evidence that commercially available AI agents can pull off complex intrusions from start to finish without human intervention. Reuters could not determine how much data, if any, was exfiltrated. Defender takeaway: exposed AI-agent infrastructure is now attacker infrastructure. Inventory any self-hosted agent platforms, lock down their network exposure, and treat autonomous-agent traffic in your environment as a new anomaly class — the operator on the other end may be asleep.

Microsoft confirmed over the weekend that this month's Windows Server security updates are causing OWA sign-in and mail-delivery failures on some on-premises Exchange Server systems. Users may be unable to sign in to Outlook on the web, messages can sit undelivered in the Outbox, and mailbox replication can fail. The root cause is a new default limit in the Windows RPC Endpoint Mapper: 512 simultaneous Endpoint Lookup calls, which is not enough for Exchange under heavy load. Affected deployments are reportedly seeing OWA fail with MapiExceptionMdbOffline (ec=1142) and Event ID 1006 in the application log. Microsoft published a registry-based workaround (raise the Endpoint Mapper limit, then restart) and explicitly urges administrators not to uninstall the Windows updates — the issue is tied to the OS update, not the separate October Exchange security updates, and any Exchange version on an impacted Windows Server can be hit. Defender takeaway: patch Tuesday now carries an availability risk for email. If you run on-premises Exchange, check OWA and mail queues before users do, apply the registry mitigation, and keep the updates installed.

Security researcher Quentin Kaiser disclosed at Hexacon in Paris a vulnerability in Broadcom's proprietary 802.11 implementation that has been present in its SDK since at least 2010. The flaw sits in the WPS EAP reassembly code: weak length checks on fragmented EAP messages allow a buffer overflow, giving an attacker within Wi-Fi range unauthenticated remote code execution on the router — no Wi-Fi password, no WPS enrollment, no user interaction. Kaiser demonstrated extracting a router's Wi-Fi password from memory over the air and obtaining a remote shell, on devices from Belgian ISPs Proximus, Orange, and VOO. He estimates roughly 1.56 million gateways exposed in Belgium alone, with the same Broadcom code in products from dozens of vendors worldwide. Broadcom released an incomplete fix in 2018, then recommended customers migrate to hostapd instead — making this a definitive "forever day" on eCos-based platforms (BCM3380, BCM3383, BCM33834). Broadcom declined to assign a CVE; CERT/CC closed its case the same day. Kaiser is releasing a fingerprinting script and a YARA rule to map global exposure. Defender takeaway: you cannot patch what has no patch. Identify Broadcom-based gateways in your estate and supply chain, disable WPS where possible, and treat ISP-provided routers as untrusted edge.

The N0n group continued its run through transport and logistics: fresh claims name Santos Brasil (port operations, Brazil), BNCT Busan New Container Terminal (South Korea), and The Akeeli Group (investigation services, US). The Storm group hit Baltimore Tank Lines, a US liquid-bulk transporter, and Emperador ransomware struck Brazil's GTL Concreto e Logística. The pattern from the weekend holds — ransomware crews are working systematically through the physical supply chain, where operational downtime converts fastest into ransom pressure. Defender takeaway: if you operate in logistics, ports, or bulk transport, assume you are in someone's target list this quarter and rehearse manual-operations continuity.

ConnectWise fallout: the SkyKick supply-chain breach from the weekend's brief is still developing — Huntress confirmed unauthorized Microsoft 365 tenant access via stolen application credentials, and ConnectWise switched off its Cloud Backup service. Tenants that ever authorized the app should still be hunting for planted persistence. Fake AI lures: ANY.RUN linked 1,258 public sandbox analyses to attackers impersonating popular AI products to distribute malware and steal browser sessions — the AI hype cycle is now a delivery vehicle. YuanKong Android RAT: a new remote-access trojan whose weakly protected dropper exposed its own operators' account identifiers while hijacking phones and stealing data.