The claim that AI might be harmless in cybersecurity just took a serious hit.
An openly available AI model, GLM 5.3, autonomously developed working exploit chains against six Firefox versions, achieving this in under 48 hours. It transformed published security patches into real-world attacks before these updates reached end users—all for roughly $1,000 in API usage fees per exploit attempt.
What sets this apart is the fully automated nature and the failure of conventional safeguards from major providers to detect or block this activity. This isn’t a proof-of-concept crafted by expert hackers; it’s an AI operating with zero human input, revealing an alarming capability gap in current defensive systems.
The crux here is the availability of open-weight models that remove barriers to entry for malicious cyber actors. The consequences extend beyond just Firefox or web browsers. It hints at a shifted cybersecurity landscape where exploits can no longer be contained behind the expertise gatekeepers once held.
We’re confronting a future where attacker automation runs at AI speeds and efficiencies, undermining traditional threat detection and patch deployment timelines.
If your security approach hasn’t started accounting for AI-generated threats, you’re already lagging behind the curve.

Leave a Reply