Markets
BTC $56,292 -8.48% ETH $1,567 -9.10% BNB $517 -8.91% XRP $0.8775 -8.78% SOL $66.11 -8.13% LINK $8.14 -7.11% BTC $56,292 -8.48% ETH $1,567 -9.10% BNB $517 -8.91% XRP $0.8775 -8.78% SOL $66.11 -8.13% LINK $8.14 -7.11% TOTAL $1.97T BTC.D 57.3%

Blockchain Technology

Google Ties a North Korean Group to Backdoored Developer Tools at Crypto Firms

Researchers link fake recruiters and tampered Claude Code settings to intrusions, plus a tool that writes exploits unattended.

State-backed hackers are targeting crypto companies in North America and Europe by turning the AI coding tools their developers already use into delivery channels for malware, according to research from Google Threat Intelligence Group and Mandiant.

The campaign, attributed to a suspected North Korean group tracked as Midnight Neptune, starts with a fake recruiter. Targets are approached on professional networking platforms by profiles using AI-generated photos and invited to online meetings, where simulated audio or video problems prompt them to run a script to fix the call, a social-engineering pattern known as ClickFix that gets the victim to execute the malicious commands personally.

Once inside, the attackers modify configuration files in the company’s internal code repositories. The target is the settings file for Anthropic’s Claude Code: the next time a developer launches the tool, the altered configuration quietly pulls down a backdoor called SOMBERMEME. The group also creates GitHub Personal Access Tokens labelled as Microsoft Copilot so the activity blends into ordinary developer traffic.

The report describes a second, more automated threat. Researchers found a working tool called exploit_agent on an exposed testing server that uses large language models to compare code before and after a security update, identify the vulnerability just patched, write an exploit and launch it, with no human in the loop. Because open-source protocol updates are public, the gap between a fix shipping and operators installing it becomes the attack window.

The tempo has picked up. Disclosed vulnerabilities and exploits nearly doubled between January and August 2026, and exploited zero-days rose from an average of eight a month in 2025 to 11 in 2026, reaching 22 in August alone.

Google also flagged LLMJacking, where criminals take over corporate cloud accounts to steal AI computing capacity and resell it through underground marketplaces it calls LLM Mixers; the fraud can exceed $100,000 in a single day and reach $1 million per victim. Separate influence networks run more than 100,000 fake social accounts to push token hype and manufacture market panic.

Google says it is deploying its own AI security tools, Big Sleep and CodeMender, to find and fix vulnerabilities before they are exploited. For crypto firms the exposure is structural: the industry runs on open-source code and small security teams, and the tools that speed up development are now a route in.

Share this article

Viber uses viber://forward?text= when the app is installed.