New Delhi: A new AI model is quietly forcing the cybersecurity industry to rethink its basics. Anthropic has introduced Claude Mythos Preview, a model that can find software flaws at a level close to top human experts. Alongside it, the company has launched Project Glasswing, a global effort to secure critical systems before these tools spread widely.
Claude Mythos, along with a new tier called “Capybara.” was leaked in late March 2026. The leak happened because of what the company called “human error” in its content system. Draft blog posts and assets were left accessible in a public data store before being taken down.
Mythos Preview has already found thousands of high-severity vulnerabilities—including some in every major operating system and web browser. pic.twitter.com/YuW484PVrr
— Anthropic (@AnthropicAI) April 7, 2026
Project Glasswing brings big tech together, even rivals
Project Glasswing brings together companies like Amazon Web Services, Apple, Google, Microsoft, Cisco, NVIDIA, and others. The goal is to use this powerful model to find and fix vulnerabilities before attackers do.
Anthropic said it created the project after seeing what its model can do. The company noted that AI has reached a point where it can “surpass all but the most skilled humans at finding and exploiting software vulnerabilities.”
More than 40 additional organisations are also part of this effort. They will use the model to scan both internal systems and open source software.
Our Project Glasswing, and so Mythos Preview, announcement is up. A lot to read in the various links, but one snippet:
“•Mythos Preview found a 27-year-old vulnerability in OpenBSD—which has a reputation as one of the most security-hardened operating systems in the world and is… https://t.co/tfgJ0HLP6g
— levent (@__alpoge__) April 7, 2026
What can Claude Mythos do?
The model has already found thousands of high-severity bugs. Some of them were sitting in code for years.
Here are a few examples shared by Anthropic:
- A 27-year-old flaw in OpenBSD that could crash systems remotely
- A 16-year-old issue in FFmpeg that testing tools missed millions of times
- Multiple bugs in the Linux kernel that allowed full system control
What stands out is that many of these were found without human help. The model worked on its own in several cases.
Anthropic CEO Dario Amodei said, “We haven’t trained it specifically to be good at cyber. We trained it to be good at code, but as a side effect of being good at code, it’s also good at cyber.”
Why does Claude Mythos raise new risks?
Claude Mythos is very powerful, but the same power can be used for attacks, and that is the concern.
Anthropic pointed out that cybercrime already costs around $500 billion every year. With AI tools like this, the barrier to finding and using vulnerabilities drops sharply.
Logan Graham from Anthropic said, “We need to prepare now for a world where these capabilities are broadly available in 6, 12, 24 months.”
What happens next?
Claude Mythos Preview for now is not public; it is rather limited to selected partners. The idea is to give companies time to strengthen their systems.
Anthropic is also investing heavily, with up to $100 million in usage credits and funding for open source security efforts.
The larger question remains. Can defenders move faster than attackers once such AI tools become common? That answer is still unclear, and the industry seems to be figuring it out in real time.