Google says autonomous AI tools can now exploit fresh patches in minutes

What happened
Google researchers reported that cybercriminals are now using autonomous AI systems to find, develop, and launch software exploits without human oversight. An exposed server running a tool called exploit_agent was found automating the process of identifying vulnerabilities, creating exploits, and carrying out network intrusions. In August 2026, attackers used this system to exploit two PaperCut print server flaws just days after the vulnerabilities were disclosed. Google is tracking several AI-driven attack frameworks that have been used against organizations worldwide.
Key details
- The exploit_agent system links to multiple cloud-based and local language models.
- In August 2026, two PaperCut vulnerabilities (CVE-2026-81578 and CVE-2026-82078) were exploited within four days of disclosure.
- Google tracks open source attack frameworks such as ARTEX, CAIRN, PentAGI, CyberStrikeAI, and HexStrike AI.
- From August to early October 2026, a customized PentAGI platform targeted many Asian organizations.
- In 2026, Mandiant handled incidents where criminals used AI to drain cloud model quotas, causing daily fraudulent bills over $100,000.
Why it matters
Security teams at organizations now face attacks that can be launched and spread at machine speed, reducing the time available to detect and respond to new threats. Companies using vulnerable software or cloud-based AI services are at greater risk of rapid, automated intrusions and financial losses. The rise of autonomous AI attack tools means defenders must adapt to faster, more frequent, and less predictable cyberattacks.
