GhostJacking and EtherHiding: New Threats Target AI Agents and Windows Users
Researchers uncover prompt-injection attacks on AI coding agents and the use of blockchain infrastructure to host uncensorable malware.
A new wave of sophisticated cyber threats is targeting the intersection of AI automation and decentralized infrastructure, creating high-success attack vectors that bypass traditional security filters. From hijacking AI coding agents via poisoned logs to hosting malware on the blockchain, these tactics represent a shift toward invisible and resilient attack infrastructure.
One of the most critical discoveries is "GhostJacking," a prompt-injection attack targeting AI coding agents such as Claude Code and Cursor. The attack works by embedding malicious instructions within security logs from providers like Cloudflare, Datadog, and Sentry. When an AI agent reads these logs to assist a developer, it inadvertently executes the embedded commands. Research unveiled at DEF CON 34 on August 9, 2026, demonstrated that GhostJacking achieved a 90% success rate in hijacking Claude Code to steal credentials and modify DNS settings.
The Rise of Decentralized Malware
Parallel to AI threats, researchers have identified "EtherHiding," a technique that leverages blockchain smart contracts to distribute malicious payloads. Specifically utilizing the Binance Smart Chain (BSC), EtherHiding allows attackers to host malware in a decentralized environment. Because the data is stored on a blockchain, these payloads are virtually impossible to remove via traditional abuse reports or domain takedowns, which typically rely on centralized hosting providers.
This decentralized approach is being paired with aggressive social engineering. The "ClickFix" campaign, first observed in April 2026, uses a fake reCAPTCHA overlay to deceive Windows users. The overlay tricks victims into pressing Win+R, Ctrl+V, and Enter, which executes a command stored in the user's clipboard. This sequence deploys GULoader malware via a UNC path, often utilizing the resilient infrastructure provided by EtherHiding to avoid detection.
Why This Shift Matters
These developments signal a dangerous evolution in the threat landscape: the exploitation of trust in AI agents and the move toward uncensorable infrastructure. GhostJacking proves that AI agents can be manipulated by the very data they are designed to analyze, turning a productivity tool into a security liability. Meanwhile, EtherHiding creates a permanent distribution network for malware that bypasses IP-based firewall rules and standard web security filters.
What to Watch
As AI agents gain deeper integration into system terminals and sensitive logs, the surface area for prompt injection will likely expand. Security teams must now consider the integrity of third-party logs as a primary attack vector. Furthermore, the use of the Binance Smart Chain for malware delivery suggests that traditional URL filtering is no longer sufficient, necessitating new methods for detecting blockchain-based payload delivery.