- Project Glasswing uses advanced AI to scan massive codebases and detect critical software vulnerabilities in seconds.s
- The new model behind Project Glasswing has already uncovered thousands of high-risk bugs across major systems.
- With support from top tech and finance companies, Project Glasswing is building a strong defense system to fix issues early.
Right now, software security is changing fast. AI is getting so good at coding that it can also break systems easily. That’s where Project Glasswing comes in. Anthropic started this project to deal with that exact problem. The idea is simple. If AI can find weaknesses, then it should be used to fix them first. Not after an attack, but before anything happens.
How does Project Glasswing actually work?
At the center of Project Glasswing is a model called Claude Mythos2 Preview. It is not publicly available yet, but the early results are kind of surprising. This model can go through millions of lines of code in a very short time.
Not just scanning, but actually understanding where things can break. It focuses on high-severity vulnerabilities, the kind that can cause real damage. In testing, it has already found thousands of serious bugs.
These were in major browsers and operating systems. That’s not a small thing. If those bugs were used by attackers, it could affect banking systems, private data, and even government-level systems.
Why is this a big deal?
Normally, security teams take time to find and fix issues. It’s manual, slow, and sometimes things get missed. But with Project Glasswing, the speed is completely different. AI does not get tired, and it does not overlook small patterns.
That’s what makes this useful right now. Because the same technology can also be used by hackers. So it’s kind of a race. Whoever uses it better wins.
Who is involved?
Anthropic is not doing this alone. Over 40 organizations are part of Project Glasswing. That includes major tech companies, infrastructure groups, and financial security firms. Some well-known names are NVIDIA, Apple, Google, the Linux Foundation, Broadcom, JPMorgan Chase, and CrowdStrike.
They are all contributing in different ways. Some provide infrastructure, some focus on security, and some help scale the system.
Anthropic is also putting money into this. Around $100 million in usage credits and about $4 million in direct funding. A big part of this is for open-source security, which is often ignored but very important.
What does this mean going forward?
The reality is simple. AI is going to be used for hacking, whether we like it or not. That part is already happening. What Project Glasswing is trying to do is stay ahead. Find the problems first, fix them early, and reduce the damage before it even starts.
It is not a perfect solution. There will still be risks. But this is probably one of the first serious attempts to use AI in a defensive way at this scale.
Overall, Project Glasswing is not just another AI project. It is more like a shift in how security is handled. Instead of reacting to attacks, the focus is now on prevention. If this works the way it is expected to, it could change how companies handle software safety. And maybe make the internet a bit more stable, which honestly is needed right now.
Keep up with the tech that actually matters.
Add us as a preferred source on Google Search for quicker access to our coverage.
