Artificial intelligence has gradually reached the stage where it can not only write code, but also systematically find errors in it faster than most people. And that's exactly what Anthropic wants to build on with its new project, which could also help society Apple.
You might be interested in
Anthropic unveiled an initiative this week ProGlasswing, which is designed to help technology companies uncover vulnerabilities in operating systems and browsers. The key tool is the Mythos model Preview, which according to the company has already found thousands of zero-day vulnerabilities across platforms.
AI as a security auditor for Apple
The first wave of partners also included AppleIn practice, this means that Apple can use this model to find errors in systems such as iOS, macOS or in a browser SafariAnd this is not a negligible change, because these layers are absolutely crucial from a security perspective. What zero-day As for vulnerabilities, these are errors that the developer is not even aware of, but an attacker can actively exploit. Once such an error appears, the race against time begins. In this regard, AI has a huge advantage. It does not get tired, it goes through millions of lines of code and looks for patterns that a person would often miss.
Nothing is 100%, not even security. Applu
Safety at Applu has long been perceived as one of the strongest points of the entire platform. But the reality of recent years shows that even a closed ecosystem is not bulletproof. With each version of the system, complexity grows and with it the room for errors. Just look at how often today Apple releases smaller updates focused purely on fixes. A typical example is the recent Update visionOS 26.3.1, which solved a specific problem without adding new features. In addition, attacks are becoming increasingly sophisticated and creative, often targeting side channels, and classic "human" testing may no longer be enough.
Attack and defense
Anthropic openly says one unpleasant thing: today's AI models can already find and exploit flaws at the level of top experts. And very soon these capabilities will expand further. This means only one thing. If AI is not used for defense, it will be used for offense. ProThe Glasswing project is essentially an attempt to "preempt" this development. The Mythos model Preview will not be publicly available and only selected partners will have access to it. In addition Appif it is about Microsoft, Google or Amazon Web Services. The financial side is also interesting. Anthropic will provide partners with credits worth up to $100 million, but further use will be paid.
Impact on users
At first glance, this may seem like a purely behind-the-scenes matter. But in reality, the impact will be felt quite quickly. If Apple If AI starts to be systematically involved in security audits, it could mean fewer critical bugs in live versions of systems and faster responses to new threats. At the same time, it is fair to say that this is a double-edged sword. The same technology that helps fix bugs can also find them for the opposite purpose. The difference will not be who has better security, but who has better AI.
I think it's already started, I'm already able to add a second birthday to a contact and add a pronoun to a name. Really from 10 to 5....