Google warns AI now powers every threat actor’s playbook

Google Alphabet Inc. ↘ 2.09% Threat Intelligence Group released its Q3 2026 AI Threat Tracker on September 8, warning that artificial intelligence has become a standard tool across all categories of cyber adversaries — from financially motivated criminals to state-linked espionage groups — and that the shift toward autonomous, agent-driven attacks is accelerating faster than many defenders can keep pace with.

"At this point, we can assume that all threat actors are using AI in some capacity and their operations have benefited," John Hultquist, chief analyst at GTIG, told Cyber Magazine. He cautioned that the agentic application of AI could create "a scaled, faster adversary".letsdatascience+1

From Prompts to Autonomous Pipelines

The report, titled "From Prompting to Autonomy: The Evolution of Adversarial AI," documents a marked shift from simple AI-assisted prompting toward multi-agent frameworks that carry out attacks with minimal human oversight. In one case, a financially motivated group compromised an organization's cloud infrastructure and assembled an autonomous framework from an AI coding chatbot, a prompt, and preconfigured markdown playbooks. The system scanned for vulnerabilities, harvested thousands of credentials, rotated IP addresses, and troubleshot errors — all in under six hours and without an operator at the keyboard.siliconangle+1

GTIG said it has not yet observed fully autonomous attack pipelines deployed against targets in the wild but noted that adversary experimentation "suggests threat actor use of AI could be evolving towards this use case".helpnetsecurity

State-Linked Actors Turning to Gemini

The tracker details how state-linked groups are embedding AI into their operations. A Russia-based adversary identified as UNC5792 used AI models to build automated monitoring bots that analyze Telegram channels for intelligence relevant to Russian authorities. Another Russian group, Sandworm (also tracked as APT44), integrated Gemini into operations targeting Ukraine, using it for social engineering, Python scripting for password spraying, endpoint fingerprinting, and projects interfacing directly with the Gemini API. An alleged China-linked espionage group attempted to use Gemini to design an automated penetration testing framework, though Google disabled the related assets before it was deployed.letsdatascience+1

Core Threats Amplified, Not Replaced

Cybersecurity practitioners broadly agree that AI has not rewritten the fundamental threat landscape so much as turbocharged it. As one InformationWeek analysis published the same day noted, "the most effective tools in threat actors' toolkits remain phishing, credential theft, impersonation and social engineering. The difference is that AI helps execute these attacks faster, at scale and with a level of polish that was previously harder to achieve".informationweek

Social engineering stands out as the area most visibly transformed. AI can now generate nearly undetectable phishing emails, realistic fake documents, and deepfake audio or video that convincingly imitates executives and business partners. Hultquist warned that speed is his foremost concern: "Criminals, like the ones who conducted a mass exploitation campaign in just six hours, will gravitate to attacks that are faster than we can respond to".siliconangle+1