Thursday, October 15, 2026

Top story

Attackers are exploiting Cisco Secure Email Gateway zero-day CVE-2026-76519 for unauthenticated root access via email. Cisco and CISA warned that CVE-2026-76519 (CVSS 9.8) was exploited before a fix: with Advanced Malware Protection File Reputation enabled and Content Scanner Tools older than 25.3.0.5525, an unauthenticated attacker can use a malicious attachment to write files and run commands as root on physical or virtual AsyncOS appliances. Cisco says fixes shipped 14 September in 15.5.5-014 and 16.5.0-780; CERT-EU notes that compromised cluster SSH keys could expose other members.

Exploitation & threats

• Public exploit code is out for FortiClient EMS CVE-2026-35616, improper access control in 7.4.5–7.4.6 (Fortinet CVSS 9.1; NVD 9.8). Fortinet (FG-IR-26-099) said exploitation was observed; 7.4.7 is the fix, the 7.2 branch is unaffected, and FortiClient Cloud and FortiSASE were already addressed.
• Palo Alto patched PAN-OS CVE-2026-0322 after limited exploitation. The management-interface flaw (CVSS 8.6) lets an authenticated attacker with network access run code as root on PA-Series, VM-Series, and Panorama. Listed source IPs are 66.42.72.14, 208.83.239.114, and 149.248.47.234; CISA set an October 17 deadline. Cloud NGFW and Prisma Access are not affected.
• CISA added Windows Shell CVE-2026-96951 (CVSS 8.8) to KEV. The October 2026 updates covered a reported 1,415 vulnerabilities, including this command-injection bug, which can bypass Mark of the Web via a crafted file or link. Microsoft confirmed active exploitation.
• Ivanti disclosed Connect Secure CVE-2026-104935 (CVSS 8.6, EPSS 26%), unauthenticated command execution on SSL VPN appliances. A limited number of customers were hit; 25.1.3.0 partially fixed it in July, but exploitation began before the backport.
• Ahsay re-patched AhsayCBS after August fixes left CVE-2026-105133 and CVE-2026-105134 chainable. Huntress saw exploitation from 7 October deploying JSP webshells and XMRig. Corrected builds are 10.3.4.45 and 9.15.2.70.
• GhostAction compromised about 1,900 GitHub accounts on 8–9 October and pushed malicious Actions workflows, disguised as security audits, at more than 30,000 repositories to steal 3,174 CI secrets.
• Microsoft dismantled YAM Store (Storm-3119), whose CoGUI, Haozi, and Iron Man System kits (Storm-2990) sent more than one billion phishing emails and texts to over 75,000 organizations since January 2026.
• Citizen Lab and Confiant described AdFection: Discover-Tech ads sent devices to Candiru spyware servers without a click. At least 45 Al Jazeera devices were linked to infections; ads also appeared on U.S. Defense Department, NATO, and IAEA networks.
• UAT-6275 (TAG-193) abused WinRAR CVE-2025-8088 from April 2026 to plant PickleShell on CIS government systems via crafted RAR archives.
• Previdian honeypots logged 1,423 attempts against Langflow CVE-2026-105740 (CVSS 9.9, authenticated RCE before 1.9.0) from 27 IPs since 19 September.

Patch priorities

• CVE-2026-76519: move Secure Email Gateway to 15.5.5-014 or 16.5.0-780 and Content Scanner Tools to 25.3.0.5525 or newer; treat exposed gateways as potentially compromised.
• CVE-2026-35616: upgrade FortiClient EMS 7.4.5/7.4.6 to 7.4.7; a public Python PoC is available.
• CVE-2026-0322: patch PAN-OS and Panorama management interfaces before the 17 October CISA date and restrict management access.
• CVE-2026-104935: confirm every supported Ivanti Connect Secure line is on a backported fix, not only 25.1.3.0.
• CVE-2026-96951: apply Microsoft’s October 2026 updates; opening a crafted file is enough to bypass Mark of the Web.
• CVE-2026-105133 / CVE-2026-105134: install AhsayCBS 10.3.4.45 or 9.15.2.70; earlier “fixed” builds still allowed webshells and miners.
• CVE-2026-55572, CVE-2026-105899, CVE-2026-106624, and CVE-2026-90532 are CVSS 10 in e107 (before 2.3.7), ColdFusion, and Adobe Experience Manager. No exploitation is reported in today’s items; patch anyway. CVE-2026-92392 (curl) and CVE-2026-109721 (Firefox 158 / ESR 153.5, Thunderbird 158) both have public PoCs.

Breaches & incidents

• UK Legal Aid Agency: April 2025 attackers stole records back to 1978, including about 250,000 pre-2007 files, with contact data, dates of birth, national ID numbers, and criminal history. Most of the trove covers applications, claims, or payments from January 2007 through 16 May 2025.
• CenterPoint Energy told Texas that 1,062,060 residents’ data was exposed, including Social Security and driver’s license numbers. A hacker claimed nearly 7.5 million records; the utility said operations were not affected.
• More than 400 European grocery stores had Danfoss AK-SM refrigeration controllers manipulated, heating units to about 50°C, spoiling food and causing small fires. Two Alsace supermarkets reported about €270,000 in losses; Spain, Portugal, Poland, Germany, and Belgium were also hit (CVE-2025-41450, CVE-2025-41451, CVE-2025-41452).
• Zscaler says an actor using the ARTEX agentic framework stole names, phone numbers, incomes, and loan limits from South Korean financial institutions via SQL injection and authentication bypass on peripheral web apps.
• Unit 42 attributes a September 2024 telecom-MSP intrusion to UNC5820 (CL-STA-0954), which used CVE-2024-47575 on FortiManager and stole configs, certificates, and keys for more than 500 FortiGates.
• Leak sites posted 29 victims in the last day, including Joliet Public Schools District 86 (Qilin), Amity University Dubai (Black Nevas), Taiwan Otsuka Pharmaceutical (Eclipse), Alma Center Servizi Medicali and Holistic Health Group (Space Bears), and FOODMatch (Akira).

AI

• Anthropic’s September threat report says suspected state-backed groups, criminals, and hacktivists used Claude Haiku, Sonnet, and Opus to automate real intrusions from December 2025 through August 2026. GTG-20006, which Anthropic says matches public reporting on Midnight Blizzard, targeted Ukrainian and European government, military, diplomatic, and defense organizations.
• Anthropic told investors Q3 revenue was $16.8 billion, about 45% above the prior quarter and more than 3.6 times 2025 annual revenue, with an 11% adjusted operating margin and a planned November IPO. It estimates 5–7% of revenue comes from distillation.
• Modal crossed $1 billion ARR and raised $750 million Series D led by Accel at a $15.75 billion valuation. Sandboxes are described as a $300 million run-rate product.
• Liquid AI released open-weight LFM2.5-24B-A2B (24B MoE, 2B active, 131,072-token context), pretrained on 33T tokens. Also on Hugging Face: Qwen-Image-2.1-Turbo, JEV-35B, and Ling-3.1-flash.
• HiddenLayer says TypeSafe AI’s Jev decision model, claimed not to hallucinate, still followed an indirect prompt injection in a shopping-agent test. UK officials have not committed to putting AI safeguards in the Cyber Security and Resilience Bill.
• NVIDIA pledged models, compute, and training to the European Commission’s AI in Science Enablers Initiative, citing Nemotron, BioNeMo, PhysicsNeMo, Earth-2, and ALCHEMI.

Watchlist

• Cisco email gateways still on old Content Scanner Tools, and any cluster whose SSH keys may have been taken.
• FortiClient EMS 7.4.5/7.4.6 now that a public PoC exists, and PAN-OS management planes ahead of 17 October.
• AhsayCBS hosts that installed the incomplete August builds, and internet-exposed Danfoss refrigeration controllers still on default credentials.
• GitHub orgs that accepted “security audit” workflow commits around 8–9 October; rotate CI secrets if so.
• Android banking malware: GhostBat RAT overlaying UPI apps in India, and Hirudo replacing Indonesian banking apps with patched copies.