The Google Threat Intelligence Group (GTIG) has published evidence of a zero-day exploit developed by a group of cybercriminals with the help of artificial intelligence (AI). It is the first time the security research group has identified a zero-day exploit created by AI and used in real-world conditions. While there is evidence that malicious actors use AI models to research and discover vulnerabilities, cases of AI-crafted zero-day exploits have proven rare or difficult to confirm.
“We observed threat actors collaborating to engineer a massive vulnerability exploitation operation,” GTIG researchers wrote in a new report on AI abuse. “Our analysis of the exploits associated with this campaign identified a zero-day vulnerability implemented in a Python script that allows a user to bypass two-factor authentication (2FA) in a popular open-source, web-based systems management tool.”
See also: Google Health is killing off the Fitbit we knew
Although GTIG has not named the affected tool, the team disclosed the vulnerability to the vendor and potentially prevented mass exploitation.
However, such incidents may become more frequent as the reasoning capabilities of AI models advance to the point where they can discover high-level logic flaws rather than just memory corruption and improper input sanitization bugs.
One such example is a recent Python 2FA bypass exploit.
“ While pioneering LLMs struggle to navigate complex enterprise authorization logic, they are now capable of performing contextual reasoning, effectively reading the programmer’s intent to correlate 2FA enforcement logic with the contradictions of its hardcoded exceptions ,” the GTIG researchers concluded . “ This ability can allow models to uncover latent logic errors that appear functionally correct to traditional scanners but are strategically flawed from a security perspective .”
GTIG has provided sufficient evidence to support that an AI model was used to both discover the vulnerability and create the exploit.
For example, the Python script contains educational strings and a fantastic CVSS score. The code also follows textbook Python programming elementsthat are consistent with the LLM educational data but a human wouldn't put into an exploit, such as detailed help menus and the clean _C ANSI color class.
See also: Hackers abuse Google Ads and Claude.ai chats for Mac attacks
Other AI-assisted vulnerability discovery elements
While the 2FA exploit was not developed through Google’s Gemini models , GTIG has discovered other instances where known threat actors have attempted to abuse Gemini for exploit discovery . This is consistent with observations from other leading AI labs such as Anthropic and OpenAI .
Google researchers recently observed a Chinese cyberespionage (UNC2814)attempting to bypass Gemini’s protections with prompts to direct the model to act as a security expert specializing in embedded devices. The attackers attempted to use such persona-driven jailbreak prompting to analyze the firmware of TP-Link and other embedded devices for vulnerabilities. Implementations of the Odette File Transfer Protocol (OFTP) were also targeted.
UNC2814 has targeted telecommunications and government agencies from more than 42 countries since 2017. The group is known to exploit vulnerabilities in edge systems and web applications.
In a separate case of AI abuse, a North Korean (APT45) was observed sending thousands of prompts to Gemini, aiming to analyze various known vulnerabilities or validate proof-of-concept exploits. The goal was likely to create a more robust arsenal of exploits for n-day vulnerabilities.
🔒 Protect your privacy with Proton VPN
Swiss VPN from the creators of Proton Mail — strict no-logs policy, strong encryption, and built-in NetShield that blocks ads, trackers, & malware.
- ✔ No-logs, based in Switzerland (except 14-Eyes)
- ✔ NetShield: blocks ads, trackers & malicious domains
- ✔ Covers all devices — free version available
The link is an affiliate link — SecNews may receive a commission at no additional cost to you. It does not affect the independence of our article writing.
See also: YouTube app for Google TV revamps side menu
“The use of these tools along with vulnerability research suggests an interest in refining AI-generated payloads in controlled environments to increase the reliability of exploits before deployment“.
The GTIG report contains other examples of AI use in the cyberattack lifecycle, including malware development and concealment, autonomous attack orchestration, infrastructure development, agentic workflows for creating deepfake content used in disinformation campaigns, and more.
