Google Confirms a New Era of Hacking: AI Agents Exploit Software Flaws Without Human Help

Google’s threat intelligence team observed multi-agent frameworks used in live intrusions and found evidence of automated exploit development

GTI
Google has uncovered AI hacking tools that can find and exploit security flaws without human intervention, raising concerns about faster cyberattacks Screenshot: @Cointelegraph/X

Google has identified cybercriminals using artificial intelligence (AI) agents to automate vulnerability discovery and exploitation without human intervention.

In a report published on 16 November 2026, its Threat Intelligence Group (GTIG) described systems capable of examining software weaknesses, developing exploits and executing them automatically, raising concerns about how quickly cyberattacks could evolve.

The findings show AI moving beyond code generation, with threat actors using interconnected agents to automate complex hacking operations that previously required significant human involvement. Google has not established that AI can independently carry out every stage of a cyberattack.

AI Agents Are Taking on More Hacking Operations

GTIG reported observing threat actors using open-source, multi-agent penetration-testing frameworks during live intrusions.

These systems can automate reconnaissance, identify software vulnerabilities, generate attack payloads and adjust their approach using feedback from previous attempts.

The frameworks identified include CAIRN, PentAGI, CyberStrikeAI and HexStrike AI. Google independently observed CAIRN and PentAGI being used in real-world operations, while CyberStrikeAI and HexStrike AI were among the tools being tested.

Between late August and early October 2026, investigators tracked a threat actor using a customised platform built with PentAGI to conduct systematic reconnaissance and vulnerability scanning against organisations across Asia.

Google also identified a suspected China-linked espionage group developing its own automated framework.

Researchers assessed that the technology could be intended for cyber espionage, although the findings do not establish direct government involvement.

Google Discovers AI Agents Automating Exploit Development

The most concerning discovery involved exploit_agent, a framework found on an exposed staging server. It used multiple advanced and locally hosted AI models to scan popular web applications for security weaknesses.

It compared software versions before and after security patches were released, analysed code changes to identify potential vulnerabilities, drafted exploits and automatically executed them.

Google said it had identified threat actors using language-model frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time.

The exposed server also contained scripts supporting post-compromise activities, including internal reconnaissance, password brute-forcing and webshell installation. Google additionally found data that had been exfiltrated from targeted victims.

Google did not confirm that exploits generated by 'exploit_agent' successfully compromised victims.

It also warned that AI-generated exploits can fail when tested, while automated frameworks may leave traces that security teams can detect.

AI Agents Could Give Hackers Less Time to Exploit Vulnerabilities

Another example involved two vulnerabilities affecting PaperCut print management servers in August 2026.

Researchers and cybersecurity firm Huntress generated working exploits within a day of disclosure, while GreyNoise subsequently reported a campaign in which an attacker likely used language models to develop and test exploits, identify targets and compromise hundreds of organisations.

The attacker reportedly escalated an initial foothold to domain administrator within minutes. However, AI involvement in that campaign remains an assessment rather than a definitively established fact.

The wider concern is the shrinking window available to organisations responding to software vulnerabilities.

Security teams must investigate weaknesses, verify patches and update exposed systems, while automated tools can potentially accelerate the process of identifying and exploiting flaws.

Google's findings do not mean human hackers are becoming unnecessary. Rather, they show how AI agents are taking over increasingly complex technical tasks, giving malicious operators the ability to conduct more automated operations with less direct involvement.