In Other News: Clop Leak Site Takeover, Docker Botnet Hunts AI Keys, Water Utility Exposure

* ShinyHunters defaced the Tor data leak site of the Cl0p ransomware gang.
* The extortion group claimed theft of server logs, source code, and private keys for Clop’s onion service.
* Researchers disclosed BragJack flaws allowing malicious extensions to control built-in AI assistants in Chrome, Edge, Opera Neon, Perplexity Comet, and the Claude extension.
* These flaws allow extensions to inject scripts or tamper with network traffic, enabling actions like reading emails, accessing local files, capturing screenshots, or activating the camera/microphone.
* Malicious versions of MemTensor’s MemOS packages were published on npm and PyPI, including a Go implant named sckit.
* The implant launches upon import/use and hunts for secrets across npm, PyPI, GitHub, AWS, and Hugging Face.
* Team Cymru found nearly 11,000 servers running Claude Relay Service or sub2api, pooling AI accounts from China and Hong Kong to reach Western frontier models.
* SpyCloud analyzed stolen identity data tied to US water and wastewater utilities, exposing credentials for remote-access systems in 1,787 organizations.
* Cisco Talos documented CLOSEDQUORUM, a Go-based implant that delegates command-and-control decisions to commercial LLMs (DeepSeek, Qwen, Mistral, Gemini).
* Canonical is replacing Ubuntu’s kernel release cycle with a single two-week cycle, aiming for faster vulnerability response.
* A flaw in TDengine (CVE-2026-42542) allows unauthenticated attackers to crash the server via malformed packets due to an integer underflow leading to a heap buffer overflow.

The convergence of AI capabilities with traditional software supply chain attacks reveals a fundamental shift in security risk assessment, moving focus from perimeter defense to the integrity of agentic systems and data flow. The emergence of attacks like BragJack demonstrates that trust relationships embedded within user-facing AI interfaces are exploitable vectors for system control, effectively turning trusted tools into remote access points. Similarly, the use of Go implants within open-source tooling signals a capability shift where software dependency management itself becomes a high-leverage attack surface, moving beyond traditional binary vulnerability scanning to inspecting runtime behaviors within agent frameworks. The AI relay networks illustrate a sophisticated method for circumventing geopolitical boundaries by leveraging decentralized infrastructure to manage access and control over large-scale computational resources, raising questions about the sovereignty of data processing versus physical location. The documented use of LLMs in command-and-control structures suggests an evolution where automated decision-making—even malicious—can be outsourced, blurring the line between human intent and machine execution. This complexity mandates a shift in defensive posture toward verifying the integrity of agent dependencies, scrutinizing the provenance of AI-generated logic, and ensuring that infrastructural controls are resilient against software-defined manipulation rather than just network intrusion.
Bridge Questions:
How should security frameworks adapt to validate the security and trustworthiness of dependencies introduced through machine learning model integration? What governance structures are necessary to manage shared access models for large-scale AI infrastructure without compromising data sovereignty? If command-and-control functions can be delegated to LLMs, what are the new accountability paradigms required when system failures or malicious actions occur?