Open-weight model builds full Firefox exploit chain in under 48 hours
An open-weight AI model called GLM-5.3 has built full Firefox exploit chains with zero human input in under 48 hours, according to a new report. The finding shows how fast autonomous AI can turn a browser’s vulnerabilities into a working attack.
The model did not need a human to find flaws, write exploit code, or connect attack steps. It produced a complete chain of Firefox vulnerabilities in less than two days.
What the report describes
The report centers on GLM-5.3, an open-weight model. Because the model’s weights are public, the result is easier to reproduce than a closed AI system would be. That makes the security warning harder to contain.
The key output was a full exploit chain, not a single bug report. The model linked multiple flaws together into one attack path. That kind of chain is normally considered advanced offensive security work.
The core warning is simple: an open-weight model can autonomously turn Firefox’s own bugs into a working exploit chain in under two days.
Key facts
- Zero human input: GLM-5.3 completed the exploit work on its own. No human researcher wrote the code or guided the attack steps.
- Full Firefox chain: The result was an end-to-end exploit chain, not a crash test or a single bug trigger.
- Under 48 hours: The model finished the entire process in less than two days, showing how quickly automated attacks can appear.
- Open-weight model: The model’s weights are publicly available. This makes the attack development method harder to restrict or track.
Why this matters
Building a working browser exploit chain has traditionally required deep expertise. The report’s experiment removes that skill barrier. If an open-weight model can do the work in 48 hours, the same approach can be repeated elsewhere.
Firefox is a widely used browser, and the demonstration is a clear example of offensive AI capability. The report does not stop at a simple proof-of-concept. It shows a model building a full attack chain with no human in the loop.
Security teams should treat AI-generated exploits as a current risk, not a future problem. The model acted autonomously, which means attack generation can be scaled once the task is set.
The threat is not a hypothetical AI. It is a model that builds Firefox exploit chains in two days.
What is an exploit chain?
An exploit chain is a sequence of vulnerabilities. Each stage of the attack gives the attacker more control. One flaw might bypass a security boundary, and the next one might allow code execution. When connected, the flaws become a complete weapon.
Single bugs are serious, but chains are more dangerous. They allow attackers to bypass multiple layers of defense. The fact that GLM-5.3 assembled this chain on its own is the central finding.
What to take away
The experiment shows that open-weight AI is strong enough for offensive security work. It also shows that speed no longer favors defenders. The attack was built in under 48 hours, with no human input.
The report positions this as a clear signal for browser security and AI oversight. The next stage of the security race will likely involve AI systems building attacks and AI systems defending against them.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.