Cyber attackers are moving beyond using GenAI for phishing emails and basic scripting, with Google Cloud researchers recently disrupting what they describe as the first zero-day exploit developed with AI.
New findings from the Google Threat Intelligence Group (GTIG) indicate that malicious actors are now experimenting with AI-assisted vulnerability discovery and exploit development, including attempts tied to initial access operations and authentication bypasses.
Google has observed “prominent cyber crime threat actors partnering to plan a mass vulnerability exploitation operation.”
The incident centered on a popular open-source web-based administration tool. Google researchers said that based on the structure and content of the exploits, they had “high confidence” that attackers used AI to identify a logic flaw that enabled them to bypass two-factor authentication (2FA) protections. They do not believe Gemini was used.
Exploit Code Clues Point to LLM Involvement
Several characteristics within the exploit code indicated that a large language model (LLM) was likely involved. According to Google’s analysis, “the script contains an abundance of educational docstrings, including a hallucinated CVSS score, and uses a structured, textbook Pythonic format highly characteristic of LLMs training data.”
The researchers said the nature of the vulnerability also pointed toward the growing strengths of frontier AI models in identifying semantic security flaws that traditional tools often miss.
The vulnerability stemmed from high-level logic error tied to a hardcoded trust assumption embedded within the application, rather than common coding weaknesses such as memory corruption or improper input sanitization.
While conventional fuzzing and static analysis tools are typically designed to detect vulnerable functions or unsafe inputs, LLMs are increasingly effective at identifying contextual inconsistencies and hardcoded anomalies, the researchers noted.
“Though frontier LLMs struggle to navigate complex enterprise authorization logic, they have an increasing ability to perform contextual reasoning, effectively reading the developer's intent to correlate the 2FA enforcement logic with the contradictions of its hardcoded exceptions.”
“This capability can allow models to surface dormant logic errors that appear functionally correct to traditional scanners but are strategically broken from a security perspective,” the researchers added.
GTIG said it worked with the vendor to fix the vulnerability before broader exploitation could take place.
In its broader analysis of adversarial AI use, GTIG said attackers are increasingly applying LLMs across the attack lifecycle, from high-speed research assistance to malware development and post-compromise activity.
“Since our February 2026 report on AI-related threat activity, Google Threat Intelligence Group (GTIG) has continued to track a maturing transition from nascent AI-enabled operations to the industrial-scale application of generative models within adversarial workflows."
For instance, the use of AI-enabled malware, such as PROMPTSPY, signals a shift toward autonomous attack orchestration, which allows threat actors to offload operational tasks to AI for scaled and adaptive activity.
“Our analysis of this malware reveals previously unreported capabilities and use cases for its integration with AI,” according to the researchers.