🚨 ACTIVE EXPLOITATION
- FakeGit Campaign Uses 7,600 GitHub Repos to Spread SmartLoader Malware — The Hacker News
Researchers uncovered a massive campaign using nearly 7,600 malicious GitHub repositories to spread SmartLoader malware.- Targets GitHub users with about 800 repos impersonating AI skills or MCP servers
- Delivers SmartLoader via copied projects, fake developer profiles, convincing README files, and malicious ZIP attachments
- SmartLoader establishes persistence and deploys StealC, an information stealer as a secondary payload
- AI agents like Anthropic Claude, Google Gemini, and OpenAI ChatGPT can be tricked into autonomously discovering and executing the malware through 'AgentBaiting'
- Over 14 million downloads recorded from roughly 200 campaign repositories as of July 2026
🕵️ RESEARCH & DEEP DIVES
Multiple vulnerabilities in FreeScout prior to 1.8.224 enable account takeover, RCE, and DoS — CVE ThreatInt
FreeScout help desk software versions before 1.8.224 contain critical vulnerabilities including account takeover and remote code execution.- CVE-2026-53595: Anonymous attackers can takeover accounts via /user-setup endpoint without authentication
- CVE-2026-53593: Authenticated remote code execution by uploading .pht web shells bypassing extension denylist
- CVE-2026-53596: Denial of service via unrestricted file upload without rate limiting
- CVE-2026-53594: Arbitrary file read through forged encrypted log file paths
- CVE-2026-53592: Prototype pollution via crafted URL query parameters in JavaScript
- All fixed in FreeScout version 1.8.224 or later
GPT-SoVITS 20250606v2pro vulnerable to OS command injection via webui.py — CVE ThreatInt
GPT-SoVITS 20250606v2pro has an OS command injection vulnerability in webui.py allowing unauthenticated attackers to execute arbitrary OS commands.- Vulnerability affects ASR, slice, denoise, and uvr5 functions
- Unsanitized Gradio textbox inputs are directly interpolated into shell commands
- Attackers inject shell metacharacters through path parameters
ktransformers before 0.6.3 vulnerable to unauthenticated pickle deserialization RCE via ZMQ — CVE ThreatInt
ktransformers versions up to 0.6.3 suffer from unauthenticated remote code execution via crafted pickle payloads sent to SchedulerServer ZMQ ROUTER socket.- Malicious reduce methods in payloads enable arbitrary shell command execution as the server process
- Fixed in commit def0f93
🔓 CVEs & KEV
- CVE-2026-44231 — CVSS 9.1 — Privilege escalation and information disclosure via REST 2.0 user collection
- CVE-2026-44230 — CVSS 6.1 — Reflected Cross-Site Scripting in search results chart
- CVE-2026-44229 — CVSS 5.4 — Cross-Site Scripting via inline-served uploaded content
- CVE-2026-55219 — Paymenter: Race condition enables credit double-spend
- CVE-2026-47198 — Paymenter: URL parameter injection bypasses paid plan limits
- CVE-2026-13381 — VSee Clinic: Insecure Direct Object Reference in File API
- CVE-2026-13380 — VSee Clinic: Exposes Cleartext SFTP Credentials in HTTP
- CVE-2026-44585 — Paymenter: Broken object level authorization via service reference manipulation
- CVE-2026-44583 — Paymenter: Blind Unauthenticated SSRF on Paypal gateway module
- CVE-2026-44584 — Paymenter: Email verification status not reset after email change
- CVE-2026-16337 — Improper authorization in ToolGroupResource and RoleAjax REST/DWR endpoints
- CVE-2026-15788 — WCOW cache mount source selector resolves NTFS junctions outside cache root