AI News

Google Finds Hackers Using AI Agents for Automated Cyberattacks

Google-Finds-Hackers-Using-AI-Agents-for-Automated-Cyberattacks

Key highlights:

  • Google says cybercriminals are now using autonomous AI frameworks like ARTEX, CAIRN, CyberStrikeAI, and PentAGI to automate various stages of their attacks.
  • The FBI has tracked ARTEX-linked attacks hitting organizations all over the world, with some hackers reaching deep access in less than three hours.
  • Hackers are turning to LLMs to analyze software patches and develop new exploits for known vulnerabilities faster.
  • AI-powered phishing services and stolen LLM credentials are helping criminals grow their operations even more.

Artificial intelligence is playing a bigger and bigger role in cybercrime. Hackers are moving on from using simple chatbots for help and are now setting up autonomous systems that can handle lots of tasks at once during an attack. Recent findings from Google Threat Intelligence Group (GTIG) and a warning from the FBI show that these technologies are making cyber intrusions quicker and easier to scale up.

With more open-source frameworks available, more kinds of criminals can use these tools. Financially motivated hackers and suspected spies are using AI-powered systems to go after organizations, exploit software flaws, and get into enterprise networks. Phishing scams and attacks on AI services are getting more automated as well.

Autonomous AI Frameworks Are Accelerating Cyber Intrusions

Google says attackers are more and more abusing open-source, agent-based penetration-testing frameworks like ARTEX, CAIRN, CyberStrikeAI, and PentAGI. These tools let hackers automate specific parts of an attack while it’s happening, so they don’t have to handle every step manually.

Instead of only using AI to write code or answer technical questions, hackers are combining different agents and tools to manage different attack tasks at the same time. Google thinks this shift means cyber operations can run at machine speeds and attack through multiple routes. Suspected China-backed spying groups and other financially driven hackers are trying out these capabilities too.

The FBI’s flash alert from November 13 points to ARTEX’s increasing popularity. The Bureau says bad actors have been using ARTEX in attacks targeting organizations worldwide since at least August 2026. The targeted industries include financial services, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecom, and legal services.

There have been attacks on South Korean financial institutions, leading to about 67,000 people’s personal data being stolen. Another attack hit ConnectWise Cloud Backup customers, giving hackers access to more than 2,250 downstream clients.

The FBI also noted how fast these attacks are happening, with some hackers moving from their first target to privileged access in under three hours. ARTEX can automate tasks like system scanning and exploiting vulnerabilities, which helps smaller groups run several campaigns at once.

ARTEX first appeared on GitHub in July 2026, posted by someone going by Autumn-27. It’s built in Go and has a Next.js web console. It connects to external AI models like Claude, GPT, DeepSeek, GLM, Grok, and Kimi. The FBI found Chinese models are most common with the attackers, especially DeepSeek Flash, since they offer strong capability, lower costs, and no visible limits during penetration testing. Even though ARTEX’s original repository was removed by its creator in October 2026, it’s still spreading in community forks like ScopeWeaver and as a commercial edition called ARTEX PRO.

LLMs Are Helping Hackers Exploit Vulnerabilities and Expand Phishing Operations

Google points out another concerning trend: hackers using LLMs to speed up vulnerability research and exploit development. More attackers are relying on AI to compare software versions, check patches, and see just how developers fixed security holes.

By studying code changes, release notes, security articles, and existing proof-of-concept code, hackers can spot weaknesses that might be left unpatched. This lets them create exploits for n-day vulnerabilities a lot faster, cutting down the time between when the flaw is made public and when it gets exploited. Google mentioned the PaperCut NG/MF vulnerabilities as a recent example.

AI is also making phishing operations simpler. Google highlighted Jieyou, a Chinese-language phishing-as-a-service platform that uses an LLM chatbot. Users can give it natural language commands to set up phishing panels and tweak templates without complex technical skills.

This way, even less experienced affiliates can customize phishing messages and pages without having to code. It shows that phishing-as-a-service providers are using AI to make their platforms more user-friendly and flexible.

Another threat is LLMjacking, where criminals grab LLM credentials and cloud resources from hacked company environments. They then resell access through underground proxy services like APIMart and JetNight, also called LLM mixers, letting others buy stolen AI resources at lower prices. All these changes show AI is supporting many parts of cybercrime, from automated intrusions and vulnerability exploitation to smarter phishing and the resale of hacked AI systems.

Devanshi Kashyap
Devanshi is a curious learner who enjoys exploring new ideas and expressing creativity through art.
You may also like
More in:AI News