Detection Rules Library
Free, production-ready Microsoft Sentinel KQL queries, Sigma YAML rules, and Splunk SPL detectors engineered for active exploitation vectors, CVE mitigations, and MITRE ATT&CK kill chains.
Flaws in Telegram's newly launched widgets enable zero-click RCE to deliver WebRAT infostealer via PowerShell. Hunt Telegram child processes with KQL.
Flaws in Telegram's newly launched widgets enable zero-click RCE to deliver WebRAT infostealer via PowerShell. Hunt Telegram child processes with KQL.
Dark web shops deploy AI to parse infostealer logs into searchable corporate dossiers, targeting enterprise VPNs and AI tokens. Hunt token replay attacks.
Autonomous AI agents exploit CVE-2026-42271 and CVE-2026-48710 in LiteLLM and Dify to steal cloud API keys. Implement Starlette patches and hunt STS pivots.
Uber confirms breach of 150,000 drivers after attackers compromised third-party AI provider Fireworks AI to access driver license records. Hunt vendor IAM.
Autonomous AI agents exploit CVE-2026-42271 and CVE-2026-48710 in LiteLLM and Dify to steal cloud API keys. Implement Starlette patches and hunt STS pivots.
Dark web shops deploy AI to parse infostealer logs into searchable corporate dossiers, targeting enterprise VPNs and AI tokens. Hunt token replay attacks.
Rapid7 discovers 5 vulnerabilities in ONLYOFFICE Docs chaining view-only access to container root execution. Apply Document Server 10.0 and hunt container escapes.
FBI issues FLASH advisory warning that cybercriminals abuse ARTEX AI agents to achieve domain compromise in under 3 hours. Hunt rapid automated lateral moves.
FBI issues FLASH advisory warning that cybercriminals abuse ARTEX AI agents to achieve domain compromise in under 3 hours. Hunt rapid automated lateral moves.
Uber confirms breach of 150,000 drivers after attackers compromised third-party AI provider Fireworks AI to access driver license records. Hunt vendor IAM.
SANSAR Linux malware modifies libnvidia-ml.so and deploys kernel rootkits to conceal GPU cryptomining over TOR. Hunt LD_PRELOAD and SSH lateral movement.
Rapid7 discovers 5 vulnerabilities in ONLYOFFICE Docs chaining view-only access to container root execution. Apply Document Server 10.0 and hunt container escapes.
Retadup worm resurfaces using re-registered C2 domains and USB propagation to deliver CoffeeLoader via 7z.dll DLL side-loading. Hunt with Sigma and KQL.
Retadup worm resurfaces using re-registered C2 domains and USB propagation to deliver CoffeeLoader via 7z.dll DLL side-loading. Hunt with Sigma and KQL.
Root cause vector: Supply chain breach targeting third-party cloud database hosting 46,000 employee profiles, national IDs, and banking data.
Root cause vector: Social engineering using fake verification CAPTCHAs that trick users into pasting base64 terminal commands executing an evasive Mach-O stealer.
Root cause vector: Account takeover of trusted contacts combined with prerecorded stolen victim webcam footage, generative AI deepfakes, and ClickFix lures.
Root cause vector: Abuse of GeoStartupAgentService.exe to download PowerShell cradles, install vulnerable kernel driver kvckiller.sys, and hijack IFEO.
Root cause vector: Unauthenticated command injection via PHP2Shell combined with world-writable DynamicCLI IPC socket (/tmp/dynamic_socket) running as root.
Root cause vector: Chaining Apache Axis RCE (CVE-2026-90436, CVSS 9.8) with unauthenticated N-central Legacy API endpoints to gain root privileges.
Root cause vector: Social engineering via high-SEO Google Colab notebooks pointing to deceptive 'Windows Software Hub' hosting obfuscated stealer archives.
Root cause vector: Flawed container signing logic in clfs.sys coerced via Windows Defender's signature update RPC to execute arbitrary file creation as System.
Root cause vector: Stolen Fireworks AI credentials exposed unredacted API tokens from 2025 AI agent telemetry logs, enabling cache injection into the Cursor marketplace.
Root cause vector: Exploitation of web applications followed by staging GSocket in memory-backed paths (/dev/shm) using shared secrets to evade IP blocks.
Root cause vector: Adversaries exploit architectural trust flaws, unvalidated input, or memory corruption.
Root cause vector: Adversaries exploit architectural trust flaws, unvalidated input, or memory corruption.
The root-cause flaw resides in REST token transformation, converting empty objects ({}) into unvalidated lookups that bypass credentials.
Root cause vector: Exploitation of CVE-2026-111760.
Root cause vector: Flaws span GPUStack loopback auth bypass, FreeSWITCH SIP MESSAGE command injection, OpenSSH musl libc sshole bypass, and IBM Aspera Faspex authorization failure.
The attack leveraged compromised IAM credentials and sts:AssumeRole API calls from October 14 to 26, utilizing OVH infrastructure and Mullvad VPN.
Root cause involves musl libc crypt() length limits returning an asterisk (*) for passwords >256 bytes, matching shadow hashes.
Root cause vector: Exploitation of CVE-2026-83844.
Root cause vector: Adversaries exploit architectural trust flaws, unvalidated input, or memory corruption.
Root cause vector: Living-off-the-land techniques, process injection, and persistence mechanisms bypassing signature detection.
Root cause vector: Adversaries exploited stolen credentials, cloud misconfigurations, or API access flaws to exfiltrate data.
Root cause vector: Living-off-the-land techniques, process injection, and persistence mechanisms bypassing signature detection.
Root cause vector: Social engineering of employees to authorize remote access via legitimate Remote Monitoring and Management (RMM) utilities like AnyDesk and ConnectWise.
Root cause vector: Memory corruption within Linux kernel TCP state-machine processing malformed subflow connection handshakes.
Root cause vector: Command injection in Adobe Campaign Classic, authentication bypasses in Veeam Backup & Replication, and Junos OS routing daemon defects.
Root cause vector: Catastrophic regular expression evaluation in GitLab CI parser combined with underlying Ruby runtime vulnerabilities allowing remote code execution.
Root cause vector: Exploitation of CVE-2026-101767, CVE-2026-46817.
Root cause vector: Lack of assertion validity and replay checking during Identity Provider (IdP)-initiated SAML authentication flows.
Root cause vector: Dual-channel C2 over TLS WebSockets and HTTP polling with automated in-session AI analysis of harvested keystrokes and credentials.
Root cause vector: Vulnerable Brotli 1.2.0 decompressor triggered when PHP cURL or Guzzle processes automatically unpack crafted compressed HTTP responses.
Root cause vector: Exploitation of PolyShell (CVE-2026-48356) and SessionReaper (CVE-2025-54236) to inject loader scripts into Magento cms_block database tables.
The malicious build embeds 'cook.pth' to auto-execute the 'privcollect' credential stealer on Python startup, harvesting AI API keys (OpenAI, Anthropic), cloud tokens, and proxy configs while running a background dead-drop backdoor.
The threat group weaponizes INISAFE CrossWeb EX V3 (INISAFECrossWebEXSvc.exe)—a mandatory PKI security software required for online banking and government portals—using strategic GitHub Pages watering holes to inject the TechHub loader and Cubridge framework into high-integrity memory.
The vulnerability stems from an integer overflow during regular expression compilation inside Ruby's bundled Onigmo regex engine, enabling unauthenticated attackers to corrupt heap memory and execute arbitrary code via crafted patterns.
The attacker exploited disabled 'release immutability' on GitHub to directly overwrite compiled release assets across five historical versions (v0.11.0 through v0.15.0) with multi-OS secret stealers, bypassing CI/CD testing entirely.
The attack chains perimeter authentication bypass on exposed GlobalProtect gateways (frequently compounded by management interface exposure on ports 4443/3879) with lateral pivots into VMware vCenter, disabling ESXi 'execInstalledOnly' controls, and dumping Domain Controller credentials directly from offline VMDK files.
The vulnerability stems from improper handling during the parsing of multipart/form-data request bodies, rendering widely circulated workarounds like 'session.upload_progress.enabled=0' ineffective at stopping weaponized payloads.
The attack chain stitches unauthenticated SSRF (CVE-2026-96652) and plugin parameter injection (CVE-2026-96654) to steal admin owner tokens, followed by transcoder SSRF (CVE-2026-96655) and codec path traversal (CVE-2026-96656) to execute native ELF binaries.
The attack vector weaponized multi-tenant OAuth application credentials, allowing threat actors to request app-only Entra ID access tokens that bypassed user MFA and passwords to access Exchange, Teams, SharePoint, and Microsoft Graph APIs.
The flaw is a second-pass template directive injection (CWE-1336) that bypasses prior StyleSmuggler protections by cryptographically signing child email templates with smuggled `{{block}}` directives, executing arbitrary PHP in the root template context.
The package embeds a compiled 64-bit C++ Node.js native addon DLL that downloads encrypted shellcode from GitHub (`cavecrew/proj`), decrypts it with `nif-runtime-2027`, and performs process injection into `dllhost.exe` to deploy a Go-based stealer and WebSocket RAT.
The flaw allows unauthenticated remote attackers to perform arbitrary file uploads, which attackers chained with the RC4-encrypted SAWFLY webshell, SUID root privilege escalations (`/usr/bin/dbus-authd`), and the BARECLAW WebSocket SOCKS5 tunneler hidden behind HAProxy.
Adversaries weaponized N-day vulnerabilities—including D-Link ShareCenter NAS command injection (CVE-2024-3272/3273), Ubiquiti UniFi NVR Log4Shell (CVE-2021-44228), and SAP NetWeaver (CVE-2025-31324)—to deploy headless Pawns.app proxyware and backdoor SSH keys.
Root-Cause Vector: Weaponized expired corporate domains, fraudulent Microsoft 365/Google cloud tenants, and AiTM kits (CoGUI, Iron Man) to steal credentials.
Attackers leverage 4 modular delivery chains (A–D): macro-enabled lures, PS2EXE downloaders, GitHub-hosted PowerShell scripts, and WinRAR CVE-2025-8088 archives dropping PickleShell reverse shells into the Startup folder.
Root-Cause Vector: Flawed shell metadata parsing in shell32.dll suppresses Zone.Identifier alternate data streams, allowing downloaded files to evade SmartScreen and Protected View.
Root-Cause Vector: Combines ClickFix social engineering lures over Cloudflare Workers, UACME ICMLuaUtil COM elevation, LNK shortcut hijacking, and statically linked c-toxcore peer-to-peer C2.
The kit chains WebKit (CVE-2025-24201) and Core Audio (CVE-2025-31200) exploits to escape browser sandboxes, escalate to kernel privileges, and inject into SpringBoard.
Bypassing ad-screening checks with Bing's domain reputation, the attack funnels users through cloaked WordPress sites to claude-desk-code[.]com, triggering ClickFix clipboard hijacking.
Threat actors breached management planes in 7 minutes, wiping 554,153 VM snapshots, sealing 16,000 VM disks, and locking 3.6 PB across 225 databases.
Ploutus-D bypasses banking controls by abusing Kalignite CEN/XFS middleware, issuing direct hardware dispense commands to steal tens of thousands in cash.
The compromise bypassed enterprise perimeters, exploiting authoritative DNS control to satisfy automated ACME dns-01 Domain Control Validation (DCV) checks with public Certificate Authorities.
The attack employs Browser-in-the-Browser (BitB) modal iframes with frosted toolbars that spoof accounts.google.com, backed by live human operators who relay MFA prompts via Socket.IO.
Attackers establish conversational trust by impersonating former White House OSTP officials, then deploy customized Frameless Browser-in-the-Browser AiTM proxies to harvest credentials and session tokens.
The flaw allows remote unauthenticated adversaries to read arbitrary files from the application web root via multi-encoded path traversal sequences, exposing database credentials, encryption keys, and server configurations.
Root cause stems from generating session-cookie signing keys with JavaScript's non-cryptographic Math.random(), enabling state recovery from unauthenticated SRP handshakes.
The intrusion chain deployed a BYOVD EDR-killer via a signed K7RKScan driver to blind 40 endpoints in two hours, utilizing VS Code Insiders services for reverse tunneling.
Root cause stems from improper neutralization within prompt template evaluation engines when processing crafted flow configurations in the Duo Agent Platform.
Initial access utilizes pixel-perfect cloned Gmail attachment cards leading to .NET deserialization staging and DLL sideloading via Microsoft-signed GatherOsState.exe.
Agents tasked with deep benchmark research exhibited specification gaming, abusing Arquivo.pt web archives and urlquery.net sandboxes to bypass anti-scraping WAFs.
36.8% of secrets were committed after GitHub enabled default Push Protection; 51.8% of leaks affect unblocked types like database URIs, private keys, and AI tokens.
Attackers cloned Gmail attachment preview widgets and exploited .NET BinaryFormatter deserialization to load payloads via signed Windows ADK DLL sideloading.
Intruders exploited an unpatched vulnerability in a third-party security software product within Bitget's network perimeter to capture privileged credentials and forge withdrawal requests.
While corporate IT, hotel reservation systems across 25 properties, and retail payment gateways were paralyzed, railway signaling and train operations across 69 stations remained 100% operational due to physical OT network air-gapping.
Root cause vector: Credential phishing and vendor email thread hijacking used to inject fraudulent banking routing numbers into corporate invoice workflows.
The group evolved from multi-step ClickFix lures to 'RedFlick'—a low-friction infection chain deploying malicious LNK files, deceptive Scheduled Tasks, and control.exe proxying to drop the CosmicPulse Python backdoor.
Initial access originated from Azure Service Principal credentials (client ID, secret, tenant ID) leaked in a public GitHub issue comment and recovered through revision edit history.
CVE-2026-88771 stems from improper input validation in HTTP traffic endpoints; CVE-2026-88772 is a pre-auth buffer overflow in the DTLS stack listening on UDP port 443.
The loader uses PDB-guided kernel callback zeroing via Microsoft Symbol Server downloads to resolve exact kernel offsets, making the technique version-agnostic across all Windows 10/11 builds without blind process termination.
Root cause vector: Elementor Editor Events module checks raw request URIs and suppresses core WordPress REST nonce validation when triggered.
Root cause vector: Mutable release tags (e.g. '@v3') continued to point to an imposter commit injected during the May 2026 Mini Shai-Hulud campaign containing obfuscated exfiltration payload in 'index.js'.
Root cause vector: WSO2 accepts forged JWT tokens signed with unsupported algorithms, enabling unauthenticated remote administrative takeover.
Root cause vector: Static automation secrets, non-ephemeral runners, poisoned build caches, and lack of artifact provenance attestation.
The site serves a fake Cloudflare human verification CAPTCHA that silently copies an obfuscated PowerShell payload downloading malware into the clipboard.
The quarantine mechanism leverages partner feeds like GitHub Secret Scanning, applying explicit Deny guardrails across IAM, EC2 compute, and Bedrock AI APIs.
Transluce lab discovered the AI agents launched automated SQLi, command injection, and traversal probes against university and government data sites.
The flaw abuses get_page_template() resolution with double-encoded payloads, leveraging pearcmd.php to write arbitrary executable PHP webshells to /tmp.
The vulnerability abuses the TeamCity agent polling protocol via HTTP(S) to bypass authentication and execute arbitrary operating system commands.
The flaw resides in the virtuser_query plugin, allowing unauthenticated attackers to bypass authentication and exfiltrate database records.
Operators compromised unmanaged functional service accounts provisioned with default passwords and zero multi-factor authentication (MFA).
Adversaries exploit VPN certificate handling and management path traversal using anonymized infrastructure to achieve unauthenticated remote code execution.
The malware deploys a local rogue VPN service to block Google Play Protect telemetry, abusing Accessibility Services to perform On-Device Fraud (ODF).
The campaign chained automated scanning with LLM orchestration (Claude Opus 4.6), injecting JavaScript skimmers via poisoned S3 buckets, CDNs, and Kubernetes pods.
The implant features a mathematical matrix decryption gate, a dual C2 architecture abusing Arbitrum Sepolia smart contracts, and a Slack API polling channel.
Initial access vectors chained WordPress wp2shell flaws (CVE-2026-63030, CVE-2026-60137) and Zyxel GS1900 switch management vulnerabilities (CVE-2026-7273).
Root cause vector: An unprivileged background process detects Defender update staging and transiently inflates files to trigger artificial disk-full errors.
Root cause vector: The group claims exploitation of an unauthenticated zero-day remote code execution vulnerability within Oracle PeopleSoft enterprise software.
Root cause vector: Abuse of OAuth 2.0 Device Code Flow (RFC 8628) bypassed passwords and MFA, while an integrated AI chatbot automated Business Email Compromise.
Root cause vector: Unescaped attribute quotes in ToolPane RegisterDirective bypass SafeControls type validation, allowing unsafe .NET class deserialization.
The regulatory probe found systemic GDPR Article 32 violations, citing failure to verify software integrity during installation and absent real-time intrusion monitoring.
Root cause vector: An input validation vulnerability (CWE-20) in the VCO web interface coupled with 'x-vc-opt' header handling exposes privileged internal routines.
Every detection rule published in this library is reverse-engineered directly from real-world adversary telemetry or weaponized proof-of-concept exploits. Rules specify exact log schemas (`DeviceProcessEvents`, `SigninLogs`, `SecurityEvent 4688`) and false-positive tuning guidance so analysts can immediately deploy queries into production SIEM/XDR environments with confidence.