Google first identified and blocked a zero-day AI exploit that bypasses two-factor authentication

Google discovered a zero‑day exploit created by AI

*Brief facts*

When and why
The Google Threat Intelligence Group (GTIG) first identified and blocked a new zero‑day exploit. The exploit was generated using artificial intelligence, not written manually by a human.

Target of the attack
A large-scale intrusion into the two‑factor authentication (2FA) system of an open‑source web application for system administration.

How AI traces were recognized
In the Python code there were “signatures” – a fake CVSS score, typical “educational” formatting and patterns characteristic of large language models (LLM).

Reason for the vulnerability
Logic error: the developer assumed that a certain component of the system was always safe and did not validate it. Through this unchecked element attackers bypassed 2FA.

Target object
Google does not disclose the specific application.

Role of Gemini
According to researchers, the Gemini model was not used in creating the exploit.

Prevention
The attack was stopped before it could become widespread.

What else GTIG noted
1. New attack methods

Attackers use AI not only to create malicious code but also to “bypass” the infrastructure of AI systems themselves.

2. Role deception

Hackers phrase prompts as if they are security experts, causing models to behave differently than intended by constraints.

3. Loading large databases

Entire vulnerability catalogs are loaded into models, and then the OpenClaw tool is used to debug malicious programs in a controlled environment before real deployment.

Conclusion
Google was able to detect and neutralize an AI‑generated exploit thanks to detailed code analysis and recognition of machine‑learning “signatures.” This underscores the need for continuous application logic checks and awareness that attackers can use AI both as a tool for creating attacks and for bypassing the protective mechanisms of systems themselves.