ConnectWise suspends Cloud Backup after credential abuse hits M365

🚨 ACTIVE EXPLOITATION

• Plex2Shell: Four-flaw Plex Media Server RCE chain exploited in the wild a day after PoC
  Attackers began exploiting the Plex2Shell vulnerability chain in outdated Plex Media Server instances one day after a public proof-of-concept was released.
  • Affects Plex Media Server before version 1.43.3.10861, especially internet-facing home and small-business Linux/Docker deployments.
  • Chains CVE-2026-96652 (SSRF), CVE-2026-96654 (URL injection), CVE-2026-96655 (transcoder SSRF), and CVE-2026-96656 (file write) to achieve unauthenticated remote code execution.
  • Observed payloads include an XMRig miner named "mylovelyminer" and a Linux reverse proxy backdoor named "myawesomeproxy".
  • Attacks came from IPs 91.92.40[.]202, 176.65.139[.]228, and 45.194.92[.]222, with payloads from 217.60.194[.]116 and mining server ota.proxystream[.]st.
  • Evidence so far comes from one researcher's honeypot compromised on October 17, 2026, and has not been confirmed as a wider campaign.
  Sources 📎 Coverage: thehackernews.com · 👁 via The Hacker News

💥 BREACHES & INCIDENTS

• ConnectWise suspends Cloud Backup after attackers abuse vendor credentials to hit Microsoft 365 tenants
  ConnectWise suspended its Cloud Backup service after attackers used a compromised vendor credential to access Microsoft 365 tenants.

  • ConnectWise Cloud Backup (formerly SkyKick) is a multitenant backup service used by MSPs for Microsoft 365 and Google Workspace data, with access to OneDrive, SharePoint, Teams and directory data
  • A compromised Microsoft 365 API secret for the ConnectWise Cloud Backup application was used to authenticate as the trusted app and access customer tenants via its existing permissions
  • Huntress confirmed unauthorized access in 162 customer tenants; attackers attempted sign-ins to over 1,000 Huntress-protected accounts and succeeded in more than 300
  • Microsoft detected broad successful principal authentication from IP 142.172.64[.]128, Microsoft Graph API reconnaissance, and attempts to add new credentials as backdoors; one tenant saw the ConnectWise Cloud Management - MS Teams app (Application ID 24205314-2b89-4e07-a2a4-579e898a5640) grant itself Global Administrator
  • ConnectWise suspended Cloud Backup, SaaS Security and Cloud Management services and disabled 7 app identities; no evidence partner backup data was compromised
  Sources 📄 Source: huntress.com · 📎 Coverage: cybersecuritydive.com · 👁 via Cybersecurity Dive

• Bullhorn takes staffing platform offline to contain cyberattack causing system-wide outage
  Bullhorn disabled access to its staffing platform after a cyberattack caused a system-wide outage.

  • Bullhorn's cloud staffing/recruitment platform serves over 10,000 customers; ATS/CRM, Automation, Back Office, Talent Platform, and sandbox environments suffered major outages, while Adapt, eEmpACT, and StaffSuite remained operational.
  • The incident began October 16, 2026 with job posting issues; Bullhorn confirmed a cyber incident on October 17 and temporarily disabled access to contain the activity, working with CrowdStrike specialists.
  • Both the Bullhorn CRM login and API were taken offline, disrupting connected services such as job board sync, candidate registrations, application submissions, and automation integrations.
  • A forensic investigation has not yet determined whether customer data was exfiltrated; all non-SSO users must reset passwords on next login and API passwords require rotation.
  • Some services began coming back online October 19 with degraded performance; no restoration timeline has been provided.
  Sources 📄 Source: bullhornplugin.com · 📎 Coverage: bleepingcomputer.com · 👁 via BleepingComputer

• OpenCode breach exposes users' AI provider API keys amid large-scale distillation attack
  OpenCode disclosed a security breach exposing API keys of 1.4% of its Console users who had external AI providers linked.

  • Affects OpenCode Console users who linked external AI providers (Anthropic, OpenAI, Google); users who received no notification email were not impacted.
  • Exposed items are third-party AI provider API keys stored in affected Console workspaces; no other data appears to have been exposed.
  • Attackers ran a large-scale distillation attack for at least 48 hours using AI agents, targeting a legacy console system OpenCode was retiring.
  • Earliest possible key exposure was October 17, 2026, 00:30 UTC; users reported unauthorized charges up to $1,700 and Google project suspensions for 'abusive activity'.
  • OpenCode's own AI models used for investigation began refusing requests due to safety safeguards; the team used OpenAI's Daybreak AI cybersecurity model for defense and is working with Google, OpenAI, and Anthropic on remediation.
  Sources 📄 Source: x.com · 📎 Coverage: bleepingcomputer.com · 👁 via BleepingComputer

🔓 CVEs & KEV

• CVE-2026-107188 — Eptura Archibus — CVSS 9.6 — Critical Unauthenticated RCE via Arbitrary JSPX Upload (CVE-2026-107188)

• CVE-2026-109437 — CognicellAI Cognition — CVSS 9.2 — Cognition before 0.13.0 Unauthenticated RCE via Local Shell BackendCognition befor...

🕵️ RESEARCH & DEEP DIVES

• (no items)

📋 ADVISORIES

• Drupal Patches Critical Field UI RCE Flaw (CVE-2026-108097) With Unauthenticated Exploit Path CVE-2026-108097
  Drupal has patched a critical remote code execution vulnerability in its core Field UI module that has exploit paths via contributed modules, including one requiring no authentication.

  • Affects Drupal core Field UI module (CVE-2026-108097, SA-CORE-2026-014, rated 19/25 critical) in versions 8.9.0+ before 10.6.19, 11.0.0 before 11.3.19, 11.4.0 before 11.4.10, and 12.0.0-alpha1 before 12.0.0-beta2
  • Not exploitable with Drupal core alone, but multiple serious exploit paths exist via contributed modules, and at least one known path requires no authentication; custom code may also expose the flaw
  • The bug let the Field UI table element's pre-render step run arbitrary PHP callables via #region_callback through call_user_func_array without validation; the flawed code existed since at least Drupal 8.9.0 (2020)
  • Fixes shipped October 14, 2026 in Drupal 11.4.10, 11.3.19, 10.6.19, and 12.0.0-beta2; Entity List module CVE-2026-108099 was patched separately in version 3.0.10
  • No public exploit code or in-the-wild attacks reported as of October 15, but Drupal warned exploits might be developed within hours or days; ~28.5% of reporting sites were on affected, unfixed branches
  Sources 📎 Coverage: securityweek.com · 👁 via SecurityWeek, Cyber Security News (+1)

• Rapid7 finds auth bypass, RCE, and database read flaws in Eptura Archibus (fixed)
  Rapid7 discovered three vulnerabilities in Eptura Archibus, including a critical unauthenticated file upload leading to remote code execution.

  • Affects Eptura Archibus, an IWMS facility management platform used by over 16,000 companies including 40% of the Fortune 500 in 2023.
  • CVE-2026-107186: a legacy API endpoint allows authentication bypass (CVSS 8.6 High), enabling full database read (CVE-2026-107187, CVSS 8.5) and unrestricted file upload (CVE-2026-107188, CVSS 9.6 Critical).
  • The insecure file upload leads to remote code execution; vulnerabilities were found during Rapid7's Vector Command continuous red teaming and are now fixed.
  • A separate path traversal flaw, CVE-2025-25652, affects Archibus 2024.03.01.109's Database Update Wizard 'Run script' and 'Server File' components, allowing access to sensitive files via ../ sequences.
  Sources 📎 Coverage: rapid7.com · 👁 via rapid7.com (discovered)

• Microsoft warns AI agents should be treated as insider risks, unveils agent security controls
  Microsoft says organizations should treat AI agents as potential insider threats and is rolling out new controls to govern them.

  • Applies to organizations deploying AI agents for tasks like coding, document review, and expense processing across connected business systems.
  • Agents can access company data, apps, and tools and act at machine speed; they may err, follow hidden instructions in documents (prompt injection), or be compromised, exposing data or altering records before humans notice.
  • Microsoft says security controls must sit outside the AI model so it cannot alter or bypass its own safeguards — CEO Satya Nadella calls this separating the supply of intelligence from the authority to act.
  • Announced expansions: Project Perception and Microsoft Agent 365 agent discovery (support for 60 local agent types), Microsoft Entra Agent ID giving agents identities separate from their human sponsors, plus Dynamic Permissions, Dynamic Agent Risk, and Dynamic Conditional Access.
  • New Guardian Agents will monitor agent activity and apply permitted safeguards at machine speed, with humans retaining control of key decisions.
  Sources 📄 Source: microsoft.com · 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)

📋 UNDER-REPORTED

• 📄 Source for Rank Math PRO supply chain attack delivered self-restoring backdoor via tainted 3.0.123 builds — rankmath.com

• 📄 Source for Rapid7 details NetScaler intrusions chaining CVE-2026-88771 and CVE-2026-88779 — support.citrix.com