Google Reports AI-Assisted Cyberattacks
Google states that hackers are leveraging AI agents to identify and exploit software vulnerabilities automatically, significantly reducing the time organizations have to apply patches.
The Google Threat Intelligence Group (GTIG) has, for the first time, detected attackers utilizing AI large language models (LLMs) to streamline software vulnerability discovery and exploitation without any human guidance.
In a recent report, they described a tool found on an exposed staging server that performed the following:
- Compared software versions
- Identified potential vulnerabilities
- Generated and executed exploit code
GTIG emphasized that in the past quarter, they identified threat actors using LLM frameworks for complete automation of vulnerability discovery and exploitation, marking a new trend in cyberattacks.
Automated Workflows and Reduced Patching Windows
The use of AI agents allows attackers to:
- Reverse engineer patches via differential analysis
- Weaponize zero-day vulnerabilities (publicly unknown security flaws)
This automation results in shorter patching windows, as organizations face increased pressure to address vulnerabilities before they are exploited.
Growing Trend of AI-Assisted Attacks
According to Google's report:
- Vulnerability disclosures more than doubled between January and August 2026.
- The average number of zero-day vulnerabilities exploited per month increased from 8 in 2025 to 11 in 2026 (up to 22 in August 2026).
These automated workflows also expedite the process of gaining domain administrator access, as demonstrated in the August 2026 PaperCut campaign, where attackers achieved this within minutes.
Multi-Agent AI Operations
AI company Anthropic, in their September report, highlighted the evolution of AI-assisted attacks into multi-agent operations, automating the entire cyberattack cycle from discovery to exploitation and data theft.
These advancements underscore the growing sophistication and scale of cyber threats in the digital era.