🕵️ RESEARCH & DEEP DIVES
Hackers Use Stolen Entra ID and Azure DevOps Credentials to Access 27 Organizations
Attackers used stolen Microsoft Entra ID app credentials and Azure DevOps identities to access at least 27 organizations in a campaign targeting MSPs.- Targets MSPs and their customers since November 1, 2026; ~45 app registrations across ~18 organizations, downstream victims in government, energy, agriculture, transportation, construction, legal, education, and nonprofit sectors
- No Microsoft vulnerability involved — the attacker used valid credentials to act as trusted applications, bypassing MFA since app-only authentication is not governed by user Conditional Access policies
- Also used an Azure DevOps federated workload identity and replayed user session tokens; one multi-tenant app reached 15 customer tenants in ~10-second loops
- Operator requests ARM tokens, then Key Vault, Microsoft Graph, and Storage tokens; 29 wildcard Graph searches followed by SharePoint downloads (~1 file/second), stealing password lists, private keys, VPN profiles, and backup credentials
- Confirmed source IP: 199.19.77.84; 95% of sign-ins were scripted, with bursts of 24 attempts in 10 seconds but pauses suggesting human control
Sources 📄 Source: todyl.com · 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)WardenStealer and Aoki infostealers steal ChatGPT, Claude, Codex tokens that survive password resets
The WardenStealer and Aoki Windows infostealers are stealing OAuth tokens, API keys, and session cookies from AI tools like ChatGPT, Claude, Codex, Cursor, and GitHub Copilot.- Targets users of local AI coding and chat tools: Claude Code, Claude Desktop, OpenAI Codex, ChatGPT desktop, Cursor, and GitHub Copilot on Windows endpoints.
- Stolen data includes OAuth access and refresh tokens, OpenAI API keys, MCP server authentication material, session cookies, and AI configuration files that often remain valid after a password reset or MFA prompt.
- SpyCloud reported over 130,000 unique WardenStealer infections since first observing the Rust-based malware-as-a-service family on August 3, 2026.
- WardenStealer spreads via ClickFix lures, malicious ads, cracked software, and fake game cheats; Aoki stores stolen AI material in a dedicated 'AISessions' collection folder.
- Stolen credentials are resold underground at steep discounts (e.g., $100 of Codex/Claude API credit for $8.94), fueling LLMjacking; IOCs include C2 domain neihite[.]cc and temp file prefix %TEMP%\akg_.
Sources 📄 Source: spycloud.com · 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)LLMJacking Attacks Leave Companies With Up to $1 Million in Fraudulent Cloud Bills
Threat actors are hijacking enterprise LLM access via stolen cloud credentials and reselling it, leaving victims with cloud bills up to $1 million.- Organizations in financial services, retail, technology, and transportation across North America, South America, and Asia Pacific were hit in incidents Mandiant investigated during 2026.
- Attackers abuse paid AI resources — unauthorized usage generated over $100,000 in charges in a single day, with cumulative bills reaching $1 million at some victims.
- Initial access comes via exposed or stolen cloud credentials (service account keys, GitHub PATs, session tokens), harvested LLM API keys from public GitHub repos, and phishing via sponsored Google Search ads leading to fake login pages.
- Group UNC6993 used a misconfigured service account key and a fake membership portal that harvested session tokens to bypass MFA; attackers validated stolen keys with tiny max_tokens=1 requests to evade detection.
- A division of labor has emerged: access brokers sell to 'LLM Mixers' like Jet Night, which resell unauthorized model access — one attacker made 500,000+ API calls in under eight hours; LiteLLM flaw CVE-2026-40217 was also exploited for gateway admin control.
AI-assisted research finds 40+ vulnerabilities in OpenEMR, fixed in 8.5.0
An AI-assisted security review found 40+ vulnerabilities in OpenEMR, all fixed in version 8.5.0.- OpenEMR, an open source EHR platform used by 100,000+ providers serving 100M+ patients in the US, India, and Indonesia, had 42 advisories credited to this research.
- Findings include SQL injection, local file disclosure, missing authorization, stored XSS, patient portal credential replacement, and administrator API impersonation.
- Severity breakdown: 3 critical, 15 high, 24 medium across 30 original reports and 12 additional finder credits.
- Researcher Fariskhi Vidyan used GPT-5.6-sol as an agent for attack surface mapping, data-flow analysis, and variant analysis, validating findings on OpenEMR 8.4.1.
- All fixes shipped in OpenEMR 8.5.0 (released 27 October 2026); CVE IDs are still pending, so advisories use GHSA IDs.
Fake software downloads use trojanized Inno installers to monetize Windows PCs
Microsoft has identified a fake-download campaign that installs monetization payloads on Windows PCs via trojanized Inno Setup installers.- Targets users searching for cracked commercial software, free games, activators, plugins, executors, and unofficial utilities on Windows devices.
- Fake installers posing as Echo Client, FabFilter Total Bundle, Navicat Premium cracks, Roblox executors, and mobile-unlock tools launch Inno Setup-style runners under %TEMP%\is-.tmp\ that stage numbered eld.exe payloads.
- A per-offer callback system contacts rotating infrastructure to select payloads, with lures delivering WorldProxyPool proxyware, CharityEngine/BOINC distributed computing, or Advanced Windows Manager utilities.
- WorldProxyPool creates three SYSTEM-level scheduled tasks (WorldProxyPoolAgent/Watchdog/Updater) to route third-party internet traffic through infected devices.
- IOCs include callback domains beliefdesk[.]xyz, companyharbor[.]xyz, signalpane[.]cfd, inkact[.]xyz, wishgrove[.]cfd, staged file %TEMP%\is-.tmp\eld.exe, and C:\Program Files\WorldProxyPool\sdk-agent.exe.
🔓 CVEs & KEV
- CVE-2026-121615 — Linksys E5600 — CVSS 9.3 — Linksys E5600 Critical Improper Authentication Enables Remote Access