Anthropic's AI model, Mythos, has sparked both excitement and worry in the tech world. The model's ability to find and exploit security holes in software has led to concerns about its potential impact on the world's digital infrastructure.
Mythos has surpassed human hackers in finding vulnerabilities in operating systems, cryptographic software, and e-commerce networks. Its ability to do so with minimal human help has raised alarms, with some experts warning of a 'world of vibe hacking' where amateurs can use AI models to find flaws and exploit them.
Security researchers are divided on the impact of AI-enabled hacking. While some, like Bruce Schneier, believe it will be a 'mess' in the short term, others think it will ultimately be good for defenders. Anthropic's Project Glasswing aims to give access to Mythos to 40 digital-infrastructure organisations to harden software before it's published.
However, the cost of AI checking is a concern, with Anthropic estimating one bug cost $20,000 to find. For software maintained by volunteers, like Linux, this could be a steep price. And with much of the world's code running on unmaintained systems, attackers could have a field day.
