Tuesday, November 17, 2026 | 04:06 PM ISTहिंदी में पढें
Business Standard
Notification Icon
userprofile IconSearch

Google flags first autonomous AI cyberattack, warns of growing threats

Google says attackers are using AI agents to identify vulnerabilities, develop exploits and automate cyber intrusions, raising concerns over faster attacks and growing security risks

artificial intelligence in cybersecurity

Google said threat actors are increasingly using open-source, multi-agent frameworks to coordinate different stages of cyber intrusions. Representational image

Sweta Kumari New Delhi

Listen to This Article

Google has flagged what it described as the first observed cyberattack in which an artificial intelligence (AI)-powered system carried out vulnerability discovery and exploitation without human intervention. The finding points to a shift in how cybercriminals use AI, from seeking help with individual tasks to deploying autonomous systems that can plan and execute multiple stages of an attack.

What are fully automated attacks

Google identified a threat actor using an AI-powered framework to automate vulnerability discovery and exploitation without human intervention. The system could examine software updates, identify weaknesses, develop exploits and execute them. It also supported follow-up activities such as network reconnaissance and attempts to maintain access to compromised systems.
 

How the first fully automated attack was identified

One of the most significant findings in the report is Google's observation, for the first time, of a threat actor using an AI-powered framework to automate vulnerability discovery and exploitation without human intervention.
 
In its report, Google identified an exposed staging server hosting a multi-agent framework called exploit_agent, which was connected to multiple advanced and locally hosted large language models (LLMs). The system was designed to continuously examine software updates, compare code released before and after security patches, identify possible vulnerabilities and develop exploits to take advantage of them.
 
Once a potential weakness was identified, the framework could automatically execute the exploit. Google also found scripts on the server that could support activities after gaining access, including internal network reconnaissance, brute-force attempts and installing webshells to maintain access to compromised systems.
 
Based on the server's activity, Google said the operator appeared to be interested in French websites associated with the media, entertainment and education sectors. The finding marks a shift from AI-assisted cyberattacks, where humans use AI to perform individual tasks, towards systems capable of independently carrying out several stages of an attack. 

AI agents are being used to automate cyberattacks

Google said threat actors are increasingly using open-source, multi-agent frameworks to coordinate different stages of cyber intrusions. These systems use multiple AI agents, each assigned specific tasks, which work together towards a common objective.
 
The report identified several frameworks being used or tested by attackers, including:
  • ARTEX
  • CAIRN
  • CyberStrikeAI
  • PentAGI
  • HexStrike AI
These tools can identify relatively simple software vulnerabilities, generate attack code based on the target's weaknesses and adjust their next steps using feedback from previous attempts. For example, CAIRN uses an approach in which agents observe an environment, assess available information, decide what to do next and execute the action.
 
Google also tracked a threat actor using a customised framework based on PentAGI between late August and early October 2026. The system was used to conduct reconnaissance and scan for vulnerabilities across organisations in Asia. The company also observed suspected espionage actors experimenting with their own automated attack frameworks.

AI speeds up vulnerability exploitation

Another concern highlighted in the report is the speed at which attackers can identify and exploit software vulnerabilities after security patches are released.
 
Typically, software companies release security updates to fix known vulnerabilities. However, attackers can examine the changes between older and updated versions of the software to understand what weakness has been fixed and develop an exploit targeting systems that have not installed the update. Google said threat actors are increasingly using LLMs and agentic frameworks to automate this process, known as differential patch analysis.
 
The report found that disclosures of vulnerabilities and their exploitation nearly doubled between January and August 2026. The average number of zero-day exploits also increased from eight per month in 2025 to 11 per month in 2026, before reaching 22 in August.
 
Google cited the exploitation of two vulnerabilities in PaperCut print management servers in August 2026. Security researchers and cybersecurity firm Huntress developed working exploits within a day of disclosure. A subsequent campaign tracked by GreyNoise involved attackers using AI-assisted methods to build and test exploits, identify targets and launch attacks against hundreds of organisations.
 
The report said one attacker was able to move from an initial foothold to domain administrator privileges within minutes. This illustrates how AI can shorten the time between identifying a weakness and gaining deeper access to an organisation's systems. 

AI aids espionage and influence campaigns

The report also highlights the use of generative AI beyond conventional cyberattacks. State-backed groups and other threat actors are using LLMs for target research, social engineering, data analysis and the creation of synthetic content.
 
Google said it has observed groups linked to Russia, China and Iran using AI to support information operations. These include generating content for different audiences, creating localised online personas and automating the distribution of political messaging across platforms.
 
In one investigation from August 2026, Google's Threat Intelligence Group (GTIG) identified a platform using the Gemini API alongside other AI services to automate an influence campaign targeting public sentiment on regional political issues, mainly across Europe and the Middle East.
 
The platform was designed to create synthetic personas, generate images and other media, and distribute content across social media platforms. Google also identified a separate system designed to manage more than 100,000 accounts across major platforms.
 
Google also flagged LLMJacking, where cybercriminals steal cloud credentials to access AI models and computing resources without authorisation. These stolen accounts and resources are sometimes sold through underground marketplaces. Mandiant investigated incidents across financial services, retail, transport and technology in 2026, with unauthorised AI usage generating fraudulent charges of over $100,000 in a single day and cumulative losses reaching $1 million in some cases.

How AI is changing the cybersecurity landscape

The use of AI in cybersecurity has been growing on both sides. Organisations are deploying AI tools to identify vulnerabilities, monitor networks and respond to suspicious activity, while cybercriminals are using the same technology to automate reconnaissance, generate malicious code and scale attacks.
 
Earlier, attackers largely used generative AI as an assistant to research targets, write phishing messages or generate code. However, the emergence of AI agents has expanded these capabilities. Such systems can use external tools, analyse results and take subsequent actions with limited human supervision.
 
Google's findings suggest that the threat is no longer limited to AI helping attackers work faster. Autonomous systems are beginning to handle multiple stages of cyber intrusions independently. However, the company noted that these tools still have limitations. Their automated activity can leave identifiable traces, making them easier to detect, and AI-generated exploits do not always work as intended.

Don't miss the most important news and views of the day. Get them on our Telegram channel

First Published: Nov 17 2026 | 4:05 PM IST