United States of America

Google: Hackers Now Use AI Agents to Attack in Hours

Is AI handing cyber attackers a decisive edge or does it finally give defenders the upper hand?

SEP 9·Updated 7 days ago

Google: Hackers Now Use AI Agents to Attack in Hours

Story

Above: Google's office in Krakow, Poland, on June 22. Image credit: Jakub Porzycki/NurPhoto/Getty Images

The Facts

Google's Threat Intelligence Group (GITG) warned that hackers and other cybercriminals are using AI in ever more sophisticated ways as part of their activities, which are increasingly targeting AI assets, in a report published Tuesday.
Google Cloud
The Chosun Daily
NBC
In one instance, GITG observed an unspecified threat actor use an AI coding chatbot, a prompt and a series of agentic instructions to plot, develop and execute a mass harvesting campaign that compromised thousands of third-party credentials in less than six hours.
Google Cloud
The unit also tracked UNC6508 — a Chinese operation allegedly targeting North American medical, academic and military research institutions since 2023 — deploying open-source AI systems inside compromised cloud environments to use their resources while avoiding detection by commercial APIs.
Google Cloud
Separately, the report said a group labelled UNC6780, also known as TeamPCP, embedded prompts referencing biological and nuclear weapons in malware loader scripts to make AI security scanners refuse or skip analysis of the malicious code beneath, among other payload delivery methods.
Google Cloud
Cyber actors across China, Russia and Iran, meanwhile, were found leveraging AI tools such as Gemini to increase the productivity of their influence operations, including using highly specific prompts to create "photorealistic" media content and refine narratives.
Google Cloud
The findings were released the same day that the NSA, CISA and the FBI published a joint advisory accusing six China-based AI companies, including DeepSeek, Moonshot AI and Alibaba, of extracting billions of tokens from U.S. frontier systems such as Claude, GPT, Gemini and Grok.
US Department of War

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

Google's findings conclusively prove that AI is widening the gap between defenders and attackers in the latter's favor. Agentic systems can probe the fence continuously at machine speed to find weaknesses, which, when they do, they exploit ruthlessly. When backed by the resources of hostile states, the threat intensifies even further, to the public's detriment.

Center for a New American Security

Infosecurity Magazine

Narrative B

While cybercriminals and hackers will use AI as a weapon, the advantage lies squarely with the defenders. Using AI as part of their arsenal, defenders can now use agents to find and patch flaws that humans never had the time or expertise to identify. In this way, defensive automation promises to build a near-impenetrable cyber wall to block the machinations of hostile actors.

Openai

Belfer Center