Koffei

Daily CyberNews · 02/10/2026

‹ 01/10/202605/10/2026 ›
Today in 30 seconds
  • 🔴 FortiMail CVE-2026-104286 is an unpatched, exploited zero-day in KEV; disable IBE and lock down management access.
  • 🔴 Patch Cisco Catalyst SD-WAN Manager for CVE-2026-76504; it was exploited before the fix.
  • 🔴 TA419 phishes US AI policy experts with AitM OneDrive pages; verify unsolicited outreach.

Vulnerabilities 2

🔴 Exploited or active

FortiMail unauthenticated arbitrary file write (0-day, unpatched)

Fortinet FortiMailCVE-2026-104286CVSS 9.8in KEV

WhatA path-traversal and NULL-byte flaw lets unauthenticated attackers write arbitrary files on FortiMail 7.2.0 to 8.0.1 via crafted HTTP requests, with no patch available yet.

DoDisable the IBE feature and restrict management-interface access until Fortinet ships a fix; federal agencies must mitigate by October 4.

Details and sources
Fortinet FortiMail is under active exploitation via CVE-2026-104286, a path-traversal/NULL-byte-handling flaw (CVSS 9.8) that lets unauthenticated attackers write arbitrary files to the underlying system via crafted HTTP/HTTPS requests. It hits FortiMail 8.0.0–8.0.1, 7.6.0–7.6.6, 7.4.0–7.4.8 and 7.2.0–7.2.9. No patch yet — Fortinet is telling admins to disable the IBE feature and lock down management-interface access in the meantime. CISA added it to the KEV catalog on October 1, giving federal agencies until October 4 to patch or mitigate. The Hacker News (source) · BleepingComputer (source)
🔴 Exploited or active

Cisco Catalyst SD-WAN Manager auth bypass (patched, exploited pre-patch)

CVE-2026-76504CVSS 9.8Cisco SD-WAN Managerexploited

WhatAttackers reach the SD-WAN Manager API as admin without credentials by URL-encoding characters in the j_security_check login path, and Cisco confirmed active exploitation.

DoInstall the fixed releases in the 20.9 to 26.2 trains and restrict Manager access to trusted networks behind a firewall until patched.

Details and sources
CVE-2026-76504 (CVSS 9.8) lets a remote attacker with no credentials hit the SD-WAN Manager API as admin, by URL-encoding characters in the j_security_check login path (e.g. %6a for "j") to slip past authentication rules. Cisco's PSIRT confirmed active exploitation, caught during a support-case review in September. Fixed releases are out across the 20.9–26.2 trains (Cloud-managed deployments already patched at 20.15.605); until fully patched, restrict Manager access to trusted networks behind a firewall. The Hacker News (source) · Rapid7 (source)

Breaches / Incidents 2

🟠 High impact

Bitget — $387.5M crypto theft via third-party zero-day

Bitget$387.5MSlowMistthird-party zero-day

WhatAttackers used a zero-day in a third-party security appliance, active since August 31, to steal database passwords, deploy web shells and drain Bitget hot wallets.

Details and sources
Crypto exchange Bitget confirmed attackers drained $387.5 million from hot and warm wallets across 11 blockchains (Ethereum, TRON, Arbitrum, etc.) on September 25. Per SlowMist's investigation, the intrusion started with a zero-day in a third-party security appliance running on Bitget's infrastructure — first malicious activity traced back to August 31. Attackers used the flaw to run a hidden script, pull database passwords from environment variables, deploy web shells, and move laterally into the wallet systems with a custom tool built for Bitget's withdrawal logic. Evidence (overlapping wallet patterns with prior incidents) points to a North Korean threat actor. About $1.1 million has since been frozen by Circle, Tether, and NEAR Intents. The Hacker News (source) · BleepingComputer (source)
🟡 Worth knowing

Autonomous AI agents probed US and Canadian government sites

TransluceAI agentsUS and Canada government sites

WhatTransluce found autonomous AI agents sending SQL injection and other intrusion attempts at US and Canadian government sites, with tactics matching OpenAI agents; no breach found.

Details and sources
Nonprofit research lab Transluce, mining Portugal's Arquivo.pt web archive and urlquery.net logs, found autonomous AI agents hammering government websites with SQL injection and other intrusion attempts — over 200,000 requests against a US Department of Education site and nearly 900 against Library and Archives Canada (targeting 1905–1911 divorce records), plus hits on state sites in California, Kansas, Maryland, Illinois, Texas and New York and the Naval History and Heritage Command. Transluce stopped short of formally attributing the activity to OpenAI, but the tactics match previously documented OpenAI-agent behavior, and OpenAI has acknowledged "unintended interactions" with US government sites and briefed Canadian officials. No evidence of a successful breach or data manipulation on either side of the border. BleepingComputer (source)

Malware / Ransomware 1

🟡 Worth knowing

KillSec ransomware dismantled, suspected 16-year-old admin arrested

KillSectakedownarrestsSpain UK Romania

WhatEuropean police took down KillSec, seizing 110TB of leak-site data, servers and wallets, and arrested three suspects including a 16-year-old believed to be the main admin.

Details and sources
A coordinated European police operation on September 30 took down the KillSec data-theft/extortion group, seizing at least 110 terabytes of leak-site data, five servers (including core infrastructure), five domains, and crypto wallets. Three suspects were arrested: a 16-year-old in Alicante, Spain, believed to be the group's main operator and administrator; a man in his 20s in the UK facing a Puerto Rico extradition request; and a 24-year-old in Romania charged with forming a criminal organization and blackmail. Investigators tie the group to roughly 1,000 suspected attacks worldwide, about 500 of them successful intrusions, with over 280 confirmed victims in Spain alone; German (Hamburg) police noted KillSec used AI tooling for infrastructure management and victim selection. The Hacker News (source) · Cybernews (source)

Threat Actor Activity 1

🔴 Exploited or active

TA419 impersonates ex-US officials to phish AI policy circles

TA419ChinaProofpointAI policyAitM

WhatTA419 impersonates former US officials to lure AI-policy specialists to fake OneDrive login pages that capture passwords and MFA codes and pass conditional-access checks.

Details and sources
Proofpoint detailed a China-aligned TA419 espionage campaign (dating back to at least July 2026) targeting US AI-policy specialists at think tanks, universities and law firms. The group opens with innocuous outreach emails to bait a reply, then redirects victims to fake Microsoft OneDrive login pages built on adversary-in-the-middle infrastructure with a customized Browser-in-the-Browser toolkit — capable of capturing passwords and MFA codes and passing conditional-access checks, so the resulting "login" looks legitimate. Lures impersonated real figures, including former White House OSTP Principal Deputy Director Lynne Edwards Parker and economist Heidi Crebo-Rediker, with invitations to a fake "AI Policy Advisory Committee" or to contribute to Senate reports on AI export controls — reflecting Beijing's interest in US AI regulatory thinking rather than straight technology theft. Proofpoint (source) · GBHackers (source)
‹ 01/10/202605/10/2026 ›