Anthropic recently introduced its new Claude Mythos Preview model, which represents a breakthrough in securing the software systems of large companies and critical infrastructure. Although this model was originally trained primarily on writing code, it very quickly turned out to be extremely capable at finding software bugs as well – and right at the level of experts.
This new artificial intelligence can analyse complex code and uncover even the most hidden software bugs entirely autonomously. What is even more important, it can connect several seemingly minor and unrelated vulnerabilities into one complex attack. The model has already successfully examined key public projects on which the internet infrastructure itself rests. Thanks to it, for example, an incredible twenty-seven-year-old bug in the OpenBSD system was discovered, as well as fundamental vulnerabilities in Linux that would have allowed attackers to easily gain complete administrator rights. All of these newly reported bugs have already been reliably fixed by the developers.
Given that such a powerful and advanced model could be misused in the wrong hands, the company decided that Claude Mythos Preview would not be widely available to the public. Instead, the company launched an initiative called Project Glasswing. Within this project, it makes the model available exclusively to selected partner organisations that manage the world's software. The main goal is to give defenders an enormous head start in finding and fixing critical bugs long before any potential attackers manage to discover them at all.
This proactive defence strategy is based on the same mechanisms that we proudly use in building the foundations of modern infrastructure within our eUrbes project. Since today's society is utterly dependent on functioning software, it is becoming clear that security will require long-term cooperation across the entire technology sector. Just as Project Glasswing demonstrates on a global scale, our technologies in the eUrbes project also engage automated mechanisms so that they themselves anticipate risks and strengthen their own resilience. We can thus confidently build a safe environment for all residents.
