Today’s roundup
AI Runs the Hack: Chinese Actor Automates Cyberattacks With DeepSeek
LLM Heist: Hijacking LiteLLM for Traffic Interception, Key Theft, and Tool-Call Injection
U.S. CISA adds a N-able N-central flaw to its Known Exploited Vulnerabilities catalog
New cPanel Critical Flaw Could Let Hosting Customers Run SQL as Database Root
Google Password Manager Attacks Could Let Malware Hijack Passkey-Protected Accounts
INC Ransomware Emerges as Dominant Actor Exploiting SonicWall SMA 1000 Flaws
Bitcoin hardware wallet maker destroys some inventory after more than $88 million stolen
31,000 Records Compromised in Breach of Liechtenstein Companies and Foundations Register
DOUBLECUP Uses ClickFix and Cached PNGs to Deliver CountLoader and DeviceManager RAT
18 Malicious npm Packages Deliver Cross-Platform RAT to Alibaba Tool Users
Summary
Unit 42 has detailed an AI-driven hacking campaign led by a Chinese actor using DeepSeek and the Hermes Agent framework. This operation autonomously scanned targets, selected exploits, and initiated attacks, demonstrating the AI's ability to perform extensive targeting analysis in minutes. Although the AI's autonomous runs did not fully breach targets due to robust configurations, manual attacks by the same actor successfully exfiltrated data from Citrix NetScaler and achieved command execution on Marimo notebook instances. The actor was traced to Zhuhai, China, and previously used AI to sort RCE bugs.
Researchers have detailed "LLM Heist," a sophisticated attack technique that exploits AI gateways like LiteLLM to intercept traffic, steal backend LLM provider keys, and inject malicious tool calls. This method leverages legitimate gateway-management functionalities, requiring initial access to an admin key or an unpatched vulnerability (e.g., CVE-2026-42271). The attack enables adversaries to reroute all LLM requests, monitor conversations, and modify responses, including forging tool calls that bypass prompt-level defenses. Mitigations include strict access controls, credential rotation, and enhanced monitoring of gateway configurations.
The U.S. CISA has added an actively exploited authentication bypass vulnerability (CVE-2026-18577, CVSS 8.2) in N-able N-central to its Known Exploited Vulnerabilities catalog. This flaw, an incomplete fix for CVE-2026-18556, allows remote attackers to gain administrative access to N-central servers and subsequently compromise managed customer endpoints. Huntress researchers observed attackers conducting reconnaissance, targeting domain controllers, and moving laterally, often using VPN exit nodes. Over half of reachable N-central cloud servers remain unpatched. CISA mandates federal agencies to remediate by August 6, 2026.
cPanel has patched a critical vulnerability, CVE-2026-58048 (CVSS 9.4), which allowed authenticated hosting customers to execute SQL commands in the database's root context. This flaw enabled privilege escalation by crossing the boundary between a cPanel account and the server's administrative database identity. The security release also addressed two other account boundary bypass routes, highlighting a significant risk to shared hosting environments.
Security researchers have uncovered new "Pass-ta-key" attacks that enable malware on already-compromised Windows devices to hijack Google Password Manager's synced passkeys. These attacks, detailed as Pass-ta-key, Silver Pass-ta-key, and Golden Pass-ta-key by Unit 42, allow malware to take over accounts, bypass user verification, and extract passkey private keys without requiring user interaction like fingerprints or PINs. The strongest attacks target the master key.
The INC Ransomware operation has been identified as a dominant threat actor actively exploiting recently disclosed security flaws in SonicWall Secure Mobile Access (SMA) 1000 series VPN appliances. Resecurity reported a significant acceleration of INC Ransomware activity since early August 2026, with multiple victims listed on its data leak site, underscoring the critical importance of patching VPN infrastructure.
The company Coinkite, maker of Coldcard hardware wallets, has announced the destruction of part of its inventory following a significant firmware vulnerability that led to the theft of over $88 million from customer Bitcoin holdings. This updates previous reports of a $70.2 million theft attributed to a March 2021 firmware error that caused deterministic seed generation.
A cyberattack on Liechtenstein's government-maintained Register of People Behind Companies and Foundations has compromised approximately 31,000 records. The exposed data includes full names, dates of birth, nationalities, countries of residence, and details of ownership or control for individuals linked to companies, foundations, and trusts. Authorities detected the intrusion, secured the data, and took the system offline, forming a crisis unit to investigate.
A new Russian loader-as-a-service, DOUBLECUP, is employing "ClickFix" attacks to conceal malicious code within PNG images cached by victims' web browsers. This technique delivers various payloads, including the CountLoader and a previously undocumented remote access trojan named DeviceManager, impacting both Windows and macOS systems. The malware leverages steganography in the first stage to retrieve and execute its hidden content.
Cybersecurity researchers have discovered 18 malicious npm packages designed to target users of Alibaba developer tools with a cross-platform remote access trojan (RAT). This sophisticated software supply chain attack primarily focuses on Chinese-speaking environments. One notable package, "lib-mtop," mimicked a legitimate private Alibaba package, indicating a targeted and deceptive approach to compromise developer systems.
Want to dig deeper?
Vulnerabilities
Malware Families