Google report: threat actors use AI to launch cyberattacks, often without human involvement
Threat actors using AI to mount cyberattacks, often without human help: Google
Google's Threat Intelligence Group released a new AI threat report, saying hackers are increasingly using LLMs to continuously scan software updates, locate newly fixed vulnerabilities and build attacks before patches are deployed, all without human involvement. The report says attackers are already using or testing offensive AI frameworks such as ARTEX, CAIRN, CyberStrikeAI and PentAGI, chaining steps such as scanning targets, discovering weaknesses and generating payloads into automated workflows; among them, the exploit_agent framework can automatically complete vulnerability analysis and exploitation without human involvement, and has been linked to a campaign targeting websites in France's media, entertainment and education sectors. The report also flagged the LLMJacking trend of reselling unauthorized access to high-performance models through underground proxy services, and said North Korea's MIDNIGHT NEPTUNE group, when attacking cryptocurrency platforms in Europe and the US, tampered with .claude/settings.json so that when users launched Claude Code it automatically downloaded the SOMBERMEME backdoor.
The report shows attackers use agents to automatically compare patches and generate exploits; security patches become high-value intelligence for attackers, and the remediation time window is greatly compressed.