Artificial intelligence (AI) is seemingly the most sought-after tool for cybercriminals and state-backed hacking groups. They are no longer limited to simple tasks such as writing phishing emails or researching targets; with agentic AI, they are taking on bigger roles.
Google, on Tuesday, September 8, introduced its latest Google Threat Intelligence Group (GTIG) report named GTIG AI Threat Tracker 2026. The report shows how attackers are increasingly relying on AI throughout the different stages of a cyberattack.
The report is based on Google’s visibility into Mandiant investigations and the misuse of Google’s Gemini AI, along with activities observed in cybercrime communities and other sources. A major highlight of the report is the growing use of AI agents, systems that can carry out multiple tasks with limited human involvement.
How are hackers using AI?
The GTIG report states that China-linked cyber espionage group BASIN CASTLE has used large language models (LLMs) for tasks that range from researching targets to troubleshooting problems at the time of an intrusion. In another case, a China-linked cyber espionage group tried to use Gemini to create an automated penetration-testing system. It was designed to perform the early stages of an intrusion with limited human intervention.
Meanwhile, Mandiant said that a large credential-theft campaign used a multi-agent AI framework running on stolen infrastructure. Google said that the campaign could compromise thousands of credentials in just six hours. These cases indicate a broader shift that attackers are moving from simply asking AI for help to using these systems to conduct a sequence of tasks and respond to problems on the fly.
New security risks
The report also identified numerous trends that emerged in the second quarter of 2026. Software supply chains are becoming more vulnerable. According to the report, AI coding tools and open-source software have made it easier and faster to develop applications. At the same time, this is also creating new opportunities for attackers who are targeting developers, AI coding assistants and security tools that rely on LLMs.
On the other hand, AI technology itself is becoming a target. Attackers are going after companies’ proprietary AI models, source code, prompts and research. Google observed such activity across industries, including government, media, healthcare, etc. Another key trend is that AI agents are becoming more capable. Cyber attackers are increasingly experimenting with systems that can automatically scan for targets, handle errors and steal credentials at scale.
The report also indicates that AI is being used in different stages of attack. Cybercriminals and state-backed groups can use these systems to research targets and generate cogent social-engineering messages, modify malware to make it harder to detect, and troubleshoot problems after accessing a system. The report also underscored another growing problem known as LLMJacking, where attackers steal AI platform accounts or hijack cloud computing resources. This is to use expensive AI services without actually paying for them.
Why does it matter?
While the report does not suggest that AI has completely automated cyberattacks, human attackers remain involved, and many AI-powered techniques are still being tested. Regardless, the report shows that AI is indeed becoming a force multiplier for attackers. This means AI can help them work faster, handle more tasks, and operate at a larger scale.
With AI systems gaining more autonomy, the cybersecurity challenge is rapidly changing. The concern is no longer what it can generate when prompted, but what attackers can make the AI systems do with limited human intervention.