Google Links Crypto Campaign to Suspected North Korean Hackers
The operators altered Claude Code settings in internal repositories to trigger a backdoor download. A separate AI framework was configured to automate vulnerability research and exploit execution.

Google's Threat Intelligence Group (GTIG) linked a crypto-company campaign in North America and Europe to suspected North Korean hackers who altered Claude Code settings to deliver a backdoor when developers opened it.
GTIG tracked the suspected cluster as MIDNIGHT NEPTUNE. The operators used fake recruiter profiles on professional networking platforms, then directed targets to fraudulent video meeting platforms and urged them to run diagnostic-repair scripts. GTIG assessed the diagnostic-repair scripts as likely written with large language models (LLMs).
After gaining initial access, operators changed .claude/settings.json files in victims' internal code repositories. Opening Claude Code then triggered a download of the SOMBERMEME backdoor. The operators also named self-generated GitHub personal access tokens to resemble Microsoft Copilot, disguising their movement through company systems.
The findings appeared in GTIG's quarterly AI threat update, published Nov. 16.
A separate finding in the update involved exploit_agent, which GTIG found on an exposed staging server. The framework linked frontier and local AI models and was configured to compare software code before and after patches, identify possible vulnerabilities, draft exploit code and execute it without human intervention.
The server held stolen data and scripts for reconnaissance, brute-force attempts and installing web backdoors. Its telemetry suggested apparent interest in French websites focused on media, entertainment and education. An example involved access-control code in version 2.19.25 of Spectra (Ultimate Addons for Gutenberg), a WordPress plugin.
GTIG's analysis found that vulnerability disclosures and exploits nearly doubled between January and August 2026. Zero-day exploits averaged 11 per month during that period, compared with eight per month in 2025, and reached 22 in August.
GTIG also observed attackers using open-source AI penetration-testing frameworks, including CAIRN and PentAGI, in intrusions. The frameworks can compress attack activity to machine speed, while reduced stealth can make it easier to detect.