Anthropic has demonstrated that an open-weight model, GLM-5.3, can autonomously build end-to-end exploits.
Understand what this means.
An intensifying arms race between attackers and defenders that benefits the companies that provide compute.
Previously the bottleneck for exploits was human expertise. That bottleneck has shifted to compute.
Nvidia is a prime example of who gains.
It plans to spend $26 billion over five years building open-weight models, and every download of an open model is demand for GPUs.
Nvidia sells the shovels while the costs of democratised capability land on everyone else.
Those costs are real. Open weights can't be recalled, and safeguards can be removed.
Meanwhile most systems in the world were written before any of this existed.
Defenders won't have a choice but to find bugs and fix them at machine speed and will have to increasingly rely AI to do so.
Attackers won't have a choice but to utilitise AI to craft exploits because vulnerabilities will close at rapid pace and exploits will require more elaborate chains of vunlerabilities to be successful.
So attackers rent GPUs, defenders have to rent GPUs, and neither can opt out.
The party that profits either way is the one selling the compute.
https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities