Google Detects Autonomous AI Cyberattack
Google has identified a cyberattack operation in which artificial intelligence independently discovered and exploited software vulnerabilities without human intervention. The findings, published by Google Threat Intelligence Group (GTIG) and Mandiant on November 16, highlight the growing use of autonomous AI systems in cybercrime.
AI System Finds and Exploits Software Vulnerabilities
Google discovered an exposed server running a multi-agent framework called exploit_agent, which was connected to several large language models.
The system could examine software updates, identify security weaknesses, develop exploits and execute them automatically.
Researchers also found tools supporting further intrusion activities, including network reconnaissance and attempts to maintain access to compromised systems.
Evidence suggested that the operator was particularly interested in French websites associated with media, entertainment and education.
Google Warns of Growing AI Cybersecurity Threats
The report identified increasing use of AI-powered hacking frameworks, including ARTEX, CAIRN, CyberStrikeAI and PentAGI.
These systems allow attackers to coordinate multiple automated tasks, reducing the time needed to identify and exploit vulnerabilities.
Google also observed automated reconnaissance targeting organisations in Asia and intrusion activity involving the US aerospace sector.
Software Exploits Nearly Double in 2026
Google reported that vulnerability disclosures and exploitation nearly doubled between January and August 2026.
Monthly zero-day exploits averaged eight in 2025 and 11 in 2026, before reaching 22 in August.
The report also highlighted financial risks from LLMJacking, where criminals steal credentials to access AI services without authorisation.
Mandiant investigated incidents in which unauthorised AI usage generated fraudulent charges exceeding $100,000 in a single day.
Google warned that increasingly autonomous attacks could make traditional cybersecurity responses more difficult, requiring organisations to strengthen automated detection and security monitoring.








