Skip to content

Image: securityweek.com · rights & removal

Executive Summary

State-sponsored and cybercrime groups are increasingly utilizing generative AI, including multi-agent offensive frameworks, to accelerate their operations. Google Threat Intelligence Group identified threat actors using an AI framework to automate vulnerability discovery and exploitation without human intervention. A specific multi-agent framework named exploitagent was discovered on an exposed staging server. This framework compares web application code against patched versions to find vulnerabilities, develop exploits, and execute them automatically. Attackers are employing frameworks like Artex, Cairn, CyberStrikeAI, and PentAGI, alongside LLMs and agentic frameworks to rapidly reverse engineer patches and weaponize vulnerabilities. Various actors have experimented with open-source tools such as Cairn and PentAGI for reconnaissance and scanning. Some groups have developed proprietary automated attack frameworks. Threat actors are also leveraging services like ZeroHollow and Agent Zero for exploit development and autonomous code execution. State-sponsored groups, including Chinese APTs and Iran-linked groups, use LLMs for content generation and infrastructure design in influence campaigns. This adoption has improved the quality of some long-running influence campaigns. The proliferation of LLM Jacking has led to marketplaces offering access to services using stolen credentials.

Facts Only

* State-sponsored and cybercrime groups are adopting generative AI, including multi-agent offensive frameworks, to increase operation speed and scale.
* Threat actors used an AI framework to automate vulnerability discovery and exploitation without human intervention.
* The multi-agent framework, exploitagent, was found on a staging server.
* exploitagent compared patched and unpatched web application code to identify vulnerabilities, develop exploits, and execute them automatically.
* The server contained stolen victim data and scripts for post-compromise activities, including reconnaissance and webshell deployment.
* Attackers use frameworks such as Artex, Cairn, CyberStrikeAI, and PentAGI to increase multi-stage intrusion velocity.
* LLMs and agentic frameworks are used to reverse engineer patches at machine speed and rapidly weaponize n-day vulnerabilities.
* Artex is the most used AI framework for threat actors, primarily Chinese-speaking groups.
* Cairn was observed being used in activity targeting the aerospace sector in the US.
* PentAGI was observed being used for systematic reconnaissance and vulnerability scanning across organizations in Asia.
* A Chinese group built its own automated attack framework.
* Tools like ZeroHollow (inference service) and Agent Zero (agent framework) are observed in use.
* State-sponsored groups, including Leaky Castle and APT42, use LLMs.
* Phishing-as-a-Service operators use LLMs for phishing panels and template modification.
* Threat actors hunt for LLM credentials to monetize via LLMJacking, leading to LLM Mixers.

Full Take

The shift toward machine-speed intrusion introduces a fundamental challenge to traditional defensive and operational paradigms. The observation that complex attack paths can be automatically chained by agents signifies a potential decoupling of high-level strategic intent from manual execution, which exacerbates the risk for critical infrastructure. While some actors are focused on stealth, other groups prioritize scale, suggesting a divergence in operational priorities driven by objectives—espionage versus purely criminal gain. The narrative suggests that the increasing use of AI is not merely an incremental change but a structural transformation where the barrier to entry for sophisticated, large-scale attacks is dramatically lowered, potentially democratizing high-level offensive capabilities among smaller entities. This creates a dual pressure: the need to defend against slower, human-centric threats while simultaneously developing defenses capable of operating at machine speed, forcing security to pivot from signature-based detection toward autonomous response systems. The underlying assumption that increased reach will compensate for reduced success rates for certain actors reflects a calculation where operational tempo is valued over absolute success probability, which must be critically examined in defensive strategy.
Bridge Questions: If the cost and expertise required for sophisticated attacks are being dramatically lowered via automation, what new definitions of "necessary defense" emerge for organizations protecting systems that cannot patch monthly? How should defenders allocate resources when balancing the need to achieve immediate stealth against the long-term goal of building resilient, autonomous detection mechanisms?
Autonomy check: No autonomy signals.

From the original · SecurityWeek

State-sponsored and cybercrime groups are increasingly adopting generative AI, including multi-agent offensive frameworks, to increase the speed and scale of their operations, Google Threat Intelligence Group (GTIG) reports. For the first time, GTIG has identified threat actors using an AI framework to automate both vulnerability discovery and exploitation without human intervention.
Read the full story at securityweek.com

Sentinel unavailable

The automated check of the source article's wording did not complete, so no result is shown.

This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.

AI Framework Autonomously Weaponized Flaws in Real Attacks, Google Says | Huntaegis