Anthropic has launched Project Glasswing alongside a powerful new AI model to help identify and fix critical software vulnerabilities before attackers can exploit them.
Quick Summary – TLDR:
- Anthropic introduces Mythos Preview, a powerful AI model with advanced coding abilities.
- Project Glasswing brings together major tech firms to secure global software systems.
- Thousands of zero day vulnerabilities already discovered, including decades old flaws.
- Experts warn AI could both strengthen and threaten cybersecurity.
What Happened?
Anthropic announced Project Glasswing, a new cybersecurity initiative involving major technology and infrastructure companies. Alongside it, the company introduced Mythos Preview, a frontier AI model capable of detecting and exploiting software vulnerabilities at a level close to top human experts.
The initiative gives select partners early access to the model so they can scan and secure critical systems before such capabilities become widely available.
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.
— Anthropic (@AnthropicAI) April 7, 2026
It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.https://t.co/NQ7IfEtYk7
The Rise of AI Powered Cybersecurity
Anthropic says its Mythos Preview model represents a major leap in AI capabilities, especially in coding and reasoning. Although it was not specifically trained for cybersecurity, its ability to understand and analyze code allows it to identify weaknesses that have gone unnoticed for years.
In recent tests, the model discovered thousands of high severity vulnerabilities, many of them classified as zero day flaws. Some of these issues were buried in widely used software systems and had remained undetected for over a decade.
This signals a turning point. AI is no longer just assisting developers. It is now capable of performing complex security analysis and even building exploit strategies on its own.
Project Glasswing and Industry Collaboration
Project Glasswing brings together a broad coalition of companies including Amazon Web Services, Apple, Microsoft, Google, Cisco, NVIDIA, Broadcom, CrowdStrike, Palo Alto Networks, and the Linux Foundation.
More than 40 additional organizations are also part of the initiative. These partners are using Mythos Preview to:
- Scan internal and open source codebases.
- Identify hidden vulnerabilities.
- Test potential exploit chains.
- Strengthen defensive cybersecurity systems.
Anthropic is also committing up to 100 million dollars in usage credits and additional funding to support open source security efforts.
The model is not being released publicly at this stage. Instead, Anthropic is following a controlled rollout to give developers time to fix issues before wider exposure.
Real World Findings Raise Urgency
Early results from Mythos Preview highlight the scale of the challenge. The AI has already uncovered vulnerabilities in:
- Major operating systems
- Popular web browsers
- Widely used software tools like FFmpeg
- Core infrastructure such as the Linux kernel
In one case, the model found a 27 year old flaw in OpenBSD, while another involved a 16 year old issue in FFmpeg that had passed millions of automated tests without detection.
The model has also shown the ability to chain multiple vulnerabilities together to gain deeper system access, a technique often used in advanced cyberattacks.
A Double Edged Sword for Security
Experts say these advancements create a new reality for cybersecurity. AI tools like Mythos can significantly improve defenses, but they also lower the barrier for attackers.
Anthropic CEO Dario Amodei noted that the model was trained to be good at coding, but its cybersecurity abilities emerged as a side effect. He also warned that more powerful models are on the way, making it critical to prepare now.
Logan Graham from Anthropic emphasized that traditional security assumptions may no longer hold as AI capabilities continue to grow rapidly.
Concerns Around Safety and Regulation
The launch comes amid growing scrutiny. Reports of leaked internal documents described Mythos as one of the most powerful models developed by Anthropic. There have also been discussions with government authorities regarding its potential risks.
In a separate incident, the company accidentally exposed a large number of source code files, raising concerns about operational security.
Regulators are increasingly focused on how such powerful AI systems should be controlled, especially when they could be used for both defense and offensive cyber operations.
SQ Magazine’s Takeaway
I think this is one of those moments where the industry clearly sees what is coming and is trying to get ahead of it. AI is about to change cybersecurity in a big way, and there is no slowing it down.
What stands out to me is how even competitors are working together here. That rarely happens unless the risk is real. If tools like Mythos fall into the wrong hands, the impact could be massive. But if used responsibly, this could finally give defenders the upper hand.
The next few months will be critical. This is not just about one model. It is about preparing for a future where AI can both break and protect the digital world at scale.
