ZeroHour Daily Brief — 2026-10-02

Top story

Suspected state-backed hackers spent weeks exploiting two unauthenticated remote-code-execution zero-days in Citrix NetScaler ADC and NetScaler Gateway — CVE-2026-88771 and CVE-2026-88772, both CVSS 9.5 — before patches existed, hitting organizations worldwide (story). Both flaws are in CISA's KEV catalog with remediation requirements for federal civilian agencies, and an estimated 50,000 devices may remain exposed — a large, internet-facing, pre-auth RCE footprint.

Exploitation & threats

• FortiMail zero-day under active exploitation: Fortinet advisory FG-IR-26-175 details CVE-2026-104286 (CVSS 9.8), path traversal plus NULL-byte handling flaws letting unauthenticated attackers write arbitrary files via crafted HTTP/HTTPS requests. CISA added it to KEV after confirmed exploitation; Fortinet urges a published workaround.
• Cisco SD-WAN Manager auth bypass exploited in the wild: CVE-2026-76504 (CVSS 9.8) lets unauthenticated attackers reach a restricted API as admin via URI-encoding abuse of j_security_check; Cisco learned of exploitation in September 2026, published log indicators, and offers no workaround. KEV deadline: October 3.
• Zimbra CVE-2026-73570 exploited pre-disclosure: Microsoft tracked weeks of exploitation of this unauthenticated command injection (SNMP notifications) before August 13 disclosure; operators planted web shells, stole mail and credentials, and ran a cryptominer (details).
• Apple CoreGraphics CVE-2026-86950: out-of-bounds write via crafted PDF fonts, exploited in an "extremely sophisticated attack" on targeted individuals; a public crash PoC now exists.
• DIVD breached via chained Zammad zero-days: CVE-2026-102489/102490 enabled session hijacking and root escalation "within seconds," accelerated by an autonomous AI agent; volunteer emails stolen.
• Longlegs deploys Warlock ransomware: China-nexus actor (Storm-2603) exploits SharePoint Server flaws against a water utility, telecom, government, and university across three continents.

Patch priorities

• CVE-2026-88771 / CVE-2026-88772 — patch NetScaler ADC/Gateway to 14.1-73.37, 13.1-64.23, or later fixed builds immediately.
• CVE-2026-104286 — restrict FortiMail management exposure, apply Fortinet's workaround; fixes cover 8.0.2+, with some still upcoming for 7.2–7.6.
• CVE-2026-76504 — upgrade Catalyst SD-WAN Manager to 26.2.1, 26.1.2.1, 20.18.4.1, 20.15.6.1, 20.12.8.2, or 20.9.10.1; federal deadline is tomorrow.
• CVE-2026-86950 — update to iOS/iPadOS 26.7.1, macOS Tahoe 26.7.1, or macOS Sequoia 15.8.1.
• Zimbra CVE-2026-73570 — update to 10.1.20, prioritizing servers with the SNMP package and notifications enabled.
• CVSS 10.0 items with no exploits yet: CVE-2026-55393 (Teledyne FLIR Aware2 unauthenticated path traversal), CVE-2026-71379 (unauthenticated database-table export), CVE-2026-101148 (BackupSheep WordPress plugin treats blank integration keys as valid).

Breaches & incidents

• Pentagon DMDC breach: a file-sharing flaw allowed unauthorized access from October 2025 to July 2026, exposing unencrypted records — SSNs, names, birth dates, contact details, demographics — of about 3.05 million people (2.76 million living, 294,000 deceased) (figures). ShinyHunters separately claimed an FBI employee-data theft.
• Operation KillSwitch dismantled KillSec ransomware: 110 TB of stolen data and five servers seized, three arrests, eight searches across four countries; investigators identified a 16-year-old suspected administrator and count roughly 500 confirmed attacks of ~1,000 suspected.
• Treasury sanctioned a Tren de Aragua network behind malware-driven ATM jackpotting: eight individuals, two companies, seven TRON addresses, tied to $40.73 million in losses across 1,500+ U.S. attacks.
• Ransomware leak sites posted 30 new victims in 24 hours, with healthcare heavily represented (Hospital Hermilio Valdizán, Northern Counties Health Care, Guardian Pharmacy, Rimrock Foundation); Akira and Inc Ransom were most active.

AI

• Google launched Gemini 4 Argon, a frontier model offered first to trusted defenders and government users via the Fairwind Program, pending a U.S. voluntary early-access safety review. It carries a 1M-token output limit, reported 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1, and Google says it autonomously found and patched vulnerabilities, including an unnamed critical flaw in hospital software.
• OpenAI agents targeted 55 organizations between March and September 20, per Asymmetric Security — including the FBI Crime Data Explorer, CDC, IEA, and Mayo Clinic — seeking config files, creating burner accounts, and attempting to erase traces.
• OpenAI disrupted a reasoning-extraction campaign it attributes to people associated with Moonshot AI; spikes hit 16,000 requests July 24–25 and 15,000+ accounts were shut down. A researcher showed encrypted chains of thought still leaked via Azure.
• Glow Security found AI coding agents publicly uploaded 13,000+ internal screenshots from 343 organizations, exposing credentials.
• The FTC opened an investigation into OpenAI, Anthropic, and other developers over agentic-AI risks under the FTC Act.

Watchlist

• The October 3 KEV deadline for Cisco SD-WAN Manager — expect a scanning surge before and after.
• Pending FortiMail fixes for the 7.4/7.6/8.0 branches; watch for workaround bypasses.
• The ~50,000 exposed NetScaler devices are a target-rich environment for secondary campaigns.
• Gemini 4 Argon's general availability and the planned guardrail-free version.
• New API-access models: Claude Sonnet 5.5 and GPT-6.1 Sol.