Generative AI Frameworks Increasingly Weaponized in Cyberattacks
Google's latest report reveals a worrisome trend: threat actors increasingly use advanced AI frameworks to autonomously discover and exploit vulnerabilities without human intervention.

Introduction: The Rise of AI in Cybercrime
AI technology is rapidly transforming various sectors, with its latest adaptation found in the world of cybercrime. State-sponsored and cybercriminal groups are now increasingly adopting generative AI and advanced offensive frameworks to enhance their operational efficiency and scale. Google Threat Intelligence Group (GTIG) recently reported significant findings regarding the use of an AI framework that automates vulnerability discovery and exploitation without any human intervention.
Exploit Agent: An Automated Framework for Threats
GTIG uncovered a multi-agent framework dubbed exploit_agent, detected on an exposed staging server. This innovative tool automates complex cybersecurity breaches by:
- Comparing patched and unpatched web application code
- Identifying vulnerabilities
- Developing exploits
- Executing attacks autonomously
The server also stored sensitive data obtained from previous victims and included scripts for post-compromise actions like reconnaissance and webshell deployment. Notably, GTIG observed activities targeting sectors such as the French media, entertainment, and education.
Adoption of Advanced Frameworks
Several frameworks are being utilized by cyber actors to enhance the speed and effectiveness of their attacks. The following frameworks have gained traction:
- Artex β Most utilized particularly among Chinese-speaking groups.
- Cairn β Seen targeting the aerospace sector in the United States.
- CybterStrikeAI
- PentAGI β Employed for systematic reconnaissance and vulnerability scanning across Asian organizations.
- HexStrike AI
Furthermore, GTIG identified a Chinese group embarking on constructing its own automated attack framework tailored for espionage.
The Dynamics of Multi-Stage Intrusions
With the integration of these AI frameworks, the dynamics of cyber intrusions have changed significantly. Cybercriminals enact multi-stage attacks at machine speed, thus dramatically scaling their operational capabilities. GTIG noted that the automated chaining of complex attack protocols is now a possibility due to these focused frameworks, marking a paradigm shift in the cyber threat landscape.
State-Sponsored Groups Optimizing AI Usage
Numerous state-sponsored groups have incorporated AI into their strategies, including:
- Leaky Castle β Linked to Chinese APTs, known for its cyber-espionage efforts.
- APT42 (Calanque Ion) β Associated with Iranian state-sponsored actors.
- Midnight Neptune (UNC1069) β A North Korean entity targeting cryptocurrency-related organizations.
These groups utilize generative AI for various malicious activities, including phishing, with the Jieyou phishing service launching a model-agnostic chatbot service to enhance the customization of phishing panels.
The Emergence of Phishing-as-a-Service (PhaaS)
The Jieyou platform, active since March 2026, has been using its functionality to help clients craft sophisticated phishing strategies by modifying templates and configurations. This development signifies the increase of PhaaS providers aligning their services with generative AI capabilities.

The Threat of LLMJacking and Automated Operations
A concerning trend has surfaced in the field of AI-related cybercrime, known as LLMJacking. Here, criminal actors seek to acquire credentials that grant access to large language models (LLMs) to exploit them commercially. This new avenue can lead to incredible financial repercussions. For example, in one instance, cumulative resource abuse associated with LLMJacking reached a staggering $1 million.
Marketplaces have sprung up to facilitate this illicit activity, selling access to various models and underlying proxy infrastructures, all functioning through pools of compromised credentials and compute instances.
Increasing Complexity and Automation in Cyberattacks
GTIG warns that the widespread adoption of advanced LLMs will contribute to the sophistication and stealth of cybercriminal operations. The findings indicate:
- Reduced human involvement will become standard practice.
- Stealth levels will continue to decline as attackers increasingly embrace the power of automated technologies.
The report emphasizes that cybercrime techniques will become more intricate. As cybersecurity measures evolve, so too will the capabilities of malicious agents.
The Perspective of Industry Experts
Industry experts are raising alarms regarding the implications of such developments for critical infrastructure. Matt Hartman, Chief Strategy Officer at Merlin, stated: βFor US critical infrastructure, the danger isnβt just more attacks, but attacks at machine speed. Utilities such as water and electricity providers can't keep up with attacks that operate at the speed of AI.β He further emphasized that such advancements lower the barrier for less sophisticated groups to engage in sophisticated cyber operations.
Key Takeaways
- Threat actors are using an AI framework called exploit_agent for automatic vulnerability discovery and exploitation.
- Frameworks like Artex and PentAGI have been identified as popular tools among attackers.
- LLMJacking is emerging as a significant concern in the cybersecurity domain, with financial losses already reported.
- Cybercriminal operations are evolving to operate at machine speed, necessitating urgent defensive shifts.
Conclusion: Preparing for an Uncertain Future
The shift towards AI-driven cyber operations represents a pivotal point in cybersecurity. As threat actors adopt increasingly sophisticated techniques powered by generative AI, organizations must take proactive measures to bolster their defenses. The warning from GTIG underscores the necessity of both solidifying cybersecurity fundamentals and innovating defensive tactics that can promptly respond to these automated threats.
Frequently Asked Questions
