AI Model GLM 5.3 Exploits Firefox Patch Gap to Breach Security

Thumbnail Image

The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.

Researchers at Epoch AI used the open-weight AI model GLM 5.3 to autonomously develop and execute exploits against six recent versions of Firefox, leveraging the patch gap before security fixes reached users. This AI-driven attack exposed users to unauthorized access and data compromise until Mozilla issued patches.[AI generated]

Why's our monitor labelling this an incident or hazard?

The article explicitly describes an AI system (GLM 5.3) being used to exploit security vulnerabilities in Firefox, leading to unauthorized access to user data, which is a direct harm to persons' privacy and security. The AI system's autonomous generation of an exploit webpage and chaining of vulnerabilities to escape sandboxing shows active use and malfunction (or misuse) of the AI system causing harm. The harm is realized, not just potential, as the exploit was demonstrated and vulnerabilities existed in released Firefox versions. Therefore, this qualifies as an AI Incident under the definitions provided, specifically harm to persons (a) and harm to communities (d).[AI generated]
AI principles
Robustness & digital securitySafety

Industries
Digital securityIT infrastructure and hosting

Affected stakeholders
Consumers

Harm types
Human or fundamental rights

AI system task:
Content generationGoal-driven organisation


Articles about this incident or hazard