Anthropic Introduces Mythos AI to Detect Critical Software Vulnerabilities

Anthropic has introduced a new initiative called Project Glasswing, aiming to strengthen software security in an era increasingly shaped by artificial intelligence. Alongside this effort, Anthropic also revealed a powerful new AI model known as Mythos Preview, which is designed to identify and address critical vulnerabilities in software systems before they can be exploited by attackers.

Project Glasswing is built around collaboration. It brings together some of the world’s largest technology and infrastructure companies, including Amazon Web Services, Apple, Microsoft, Google, and NVIDIA. By working together, these organizations aim to secure both proprietary and open-source software that underpins global digital systems. The goal is to detect weaknesses early and fix them before they become real threats.

Anthropic Introduces Mythos AI to Detect Critical Software Vulnerabilities

At the center of this initiative is Mythos Preview, an advanced AI model with strong coding and analytical abilities. Although it was not specifically trained for cybersecurity tasks, it has shown an impressive capacity to examine code and uncover hidden flaws. In early testing, the model discovered thousands of serious vulnerabilities, including many “zero-day” issues. These are flaws that are unknown to developers and can be exploited by attackers without warning.

Some of the vulnerabilities identified by the AI were especially surprising. In certain cases, the model found issues that had existed in widely used software for many years. For example, it detected long-standing flaws in core systems such as the Linux kernel and common tools like FFmpeg. One vulnerability had reportedly remained hidden for over two decades, highlighting how even mature software can contain unnoticed risks.

The capabilities of Mythos Preview go beyond simple detection. The model can also analyze how different vulnerabilities might be combined to create more complex attack paths. This technique, often used by skilled hackers, allows attackers to gain deeper access to systems. By identifying these chains in advance, developers can better protect their software against sophisticated threats.

See Also: Microsoft Supports Anthropic in Legal Fight Over Pentagon Ban

Despite these benefits, the technology also raises concerns. Experts warn that tools like Mythos Preview could be used for harmful purposes if they fall into the wrong hands. While they can strengthen defenses, they may also lower the barrier for cyberattacks by making advanced techniques more accessible. This creates a challenging balance between innovation and security.

To manage these risks, Anthropic is not releasing the model to the public at this stage. Instead, it is providing controlled access to selected partners, giving them time to address vulnerabilities before broader exposure. The company has also committed significant funding to support open-source security efforts, recognizing the importance of protecting shared digital infrastructure.

The launch of Project Glasswing reflects a larger shift in cybersecurity. Artificial intelligence is no longer just a support tool; it is becoming a central player in identifying and preventing threats. As AI systems continue to evolve, initiatives like this may play a key role in shaping a safer digital future, while also prompting important discussions about regulation and responsible use.