Anthropic’s Secretive New Model and Its Transparency Play

In what has undoubtedly shaped as a rapidly intensifying AI race, Anthropic has introduced “Glasswing,” a project based on their new Claude Mythos Preview model and one that signals both ambition and restraint.

The initiative is positioned as a leap forward in understanding advanced AI systems and is far from an isolated research project – it’s tied to a brand new model the company is intentionally keeping out of public hands.

A Model You Can’t Use – And That’s the Point

In a world where there’s a steady stream of increasingly accessible artificial intelligence-based tools, Anthropic’s latest Claude Mythos Preview model is notably not being released for general public use.

Wrapped under the wing of project Glasswing, the initiative will be available to select tech giants like Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks in what Anthropic describes as “an effort to secure the world’s most critical software.”

The decision to avoid a market-wide rollout underscores a growing tension in the software industry – balancing innovation with safety.

Anthropic appears to be taking a very cautious approach, prioritizing deep analysis over rapid deployment. The reasons for this were outlined by the team themselves:

Mythos Preview has already found thousands of high-severity vulnerabilities – including some in every major operating system and web browser.

The model itself is being described as highly advanced and designed not just for performance but for interpretability. In simpler terms, it’s built to be studied as much as it is to be used. In fact, Anthropic is committing up to $100 million in Mythos Preview usage credits for the project’s partners and promises to report back what they learn.

You may also like:

Inside the Black Box

At the heart of project Glasswing is the aim to identify vulnerabilities and exploits using Claude Mythos Preview.

Anthropic revealed that many of the thousands of bugs the model already detected were critical and found in virtually every major operating system and every major web browser, along with a multitude of other important software. These were zero-day vulnerabilities, meaning that the flaws were previously unknown to software developers.

For instance:

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 used to run firewalls and other critical infrastructure. The vulnerability allowed an attacker to remotely crash any machine running the operating system just by connecting to it.

In this sense, Glasswing represents an important shift in how AI development is currently approached. Rather than racing to release the most capable systems, Anthropic is investing in understanding them first.