Anthropic has launched Project Glasswing, a cybersecurity programme that unites leading companies such as Amazon Web Services, Apple and Microsoft to secure critical software infrastructure using advanced artificial intelligence (AI) technologies.
The initiative is driven by Anthropic’s unreleased model, Claude Mythos Preview, which can spot software vulnerabilities early and help security teams fix them in time, reducing the chances of attackers exploiting these weaknesses.
Claude Mythos Preview scans software code and systems to find hidden security flaws, including critical vulnerabilities that may have gone unnoticed for years across widely used platforms and digital infrastructure, helping companies identify risks earlier.
The model can also show how hackers could exploit these weaknesses, helping organisations understand the risks better and take early action to fix issues before they turn into real cyberattacks.
Anthropic said the model has already detected thousands of high-severity vulnerabilities across operating systems, web browsers and widely used software, with these findings responsibly disclosed so developers can address and fix them quickly.
According to the company, it does not plan to release Claude Mythos Preview publicly for now but aims to enable safe use of similar models in the future, with strong safeguards in place to prevent misuse of such advanced AI systems.
The company said it plans to roll out new safety measures with its upcoming Claude Opus model, helping refine systems that can detect and block harmful outputs without carrying the same level of risk as Mythos Preview.
For everyday users, the initiative signals better digital safety as AI tools can help companies detect and fix hidden software flaws faster, reducing the risk of data breaches, service disruptions and cyberattacks across essential services.
According to the company, Project Glasswing includes over 40 organisations, with up to $100mn in credits and funding committed to support open-source security and improve cybersecurity standards as AI-driven threats evolve.
