Breaking: AI-Powered Zero-Day Exploit Discovered as Adversaries Industrialize Generative Model Use

Google Threat Intelligence Group (GTIG) has uncovered the first known instance of a threat actor using a zero-day exploit believed to be developed with artificial intelligence. The exploit was intended for a mass exploitation campaign but may have been neutralized through proactive countermeasures. This marks a critical escalation in how adversaries are weaponizing generative AI for cyberattacks.

According to GTIG's latest report, a criminal threat actor had planned to use the AI-generated exploit against widespread vulnerabilities. We believe this is the first time we've seen a zero-day developed with AI, said a senior threat analyst at Mandiant, which contributed to the investigation. Our early detection likely prevented a large-scale breach.

Key Findings

Vulnerability Discovery and Exploit Generation

State-sponsored groups associated with China (PRC) and North Korea (DPRK) have also shown strong interest in using AI for vulnerability research. GTIG found evidence of these adversaries experimenting with AI to automate the discovery of new software flaws.

AI-Augmented Development for Defense Evasion

Adversaries are using AI-driven coding to build obfuscation networks and polymorphic malware. Suspected Russia-nexus threat actors have integrated AI-generated decoy logic into their malware to evade detection, making attacks harder to trace.

Autonomous Malware Operations

The emergence of AI-enabled malware like PROMPTSPY signals a shift toward autonomous attack orchestration. This malware interprets system states to generate commands and manipulate victim environments without human intervention. AI is offloading operational tasks, enabling adaptive and scaled attacks, the report notes.

AI-Augmented Research and Information Operations

Adversaries are using AI as a high-speed research assistant across the attack lifecycle. In information operations, such as the pro-Russia campaign Operation Overload, AI generates synthetic media and deepfake content to fabricate digital consensus.

Obfuscated LLM Access

Threat actors are using professionalized middleware and automated registration pipelines to obtain anonymized, premium-tier access to large language models (LLMs). This infrastructure bypasses usage limits and subsidizes attacks through trial abuse and account cycling.

Supply Chain Attacks on AI Environments

Groups like TeamPCP (UNC6780) are targeting AI software dependencies as an initial access vector. These supply chain attacks can lead to multiple compromise scenarios, from data theft to operational disruption.

Background

This report updates GTIG's February 2026 findings on AI-related threats. The shift from nascent AI-enabled operations to industrial-scale application has accelerated over the past year. Mandiant incident response engagements, Gemini analysis, and GTIG proactive research contributed to the insights. The dual nature of the threat—AI as a sophisticated attack engine and as a high-value target itself—dominates the current landscape.

Google's threat intelligence team emphasizes that these developments are not hypothetical. Attackers are actively deploying these techniques in real-world campaigns, and the speed of evolution outpaces traditional defense measures.

What This Means

The industrial-scale use of generative AI lowers barriers for sophisticated attacks, making them more accessible to a broader range of threat actors. The discovery of an AI-generated zero-day exploit is a watershed moment: it proves that AI can independently craft novel vulnerabilities for mass exploitation.

Organizations must prepare for rapid, adaptive malware that uses AI to adjust tactics in real time. Defenders need to invest in AI-powered detection and adopt proactive countermeasures, such as AI-specific threat hunting and supply chain security for machine learning dependencies. The era of AI-driven cyber warfare has arrived, warned a GTIG researcher. Attackers are moving faster than ever, and our defenses must evolve at the same pace.