Project Glasswing: Anthropic Debuts Claude Mythos Preview Its First Cybersecurity-Focused AI Model

Tuesday, Anthropic announced the launch of Project Glasswing, a new program focused on artificial intelligence (AI)-powered cybersecurity that is being developed in an effort called Project glasswing. This Claude Mythos Preview is general-purpose model of powering the project to be run by the San Francisco-based AI firm. The company describes the model as its most capable for coding and agentic work. Only Project Glasswing partners can now see security flaws and vulnerabilities across their tech stack, it is currently available. It has also spent money on the initiative by the AI firm, which is also a funder for .

Anthropic’s Project Glasswing is Here

A post on X (formerly Twitter) announced the initiative, with launch partners Amazon Web Services (AWS), Apple, Broadcom, Cisco and CrowdStrike; Google/JPMorganChase; the Linux Foundation Microsoft Nvidia — Palo Alto and its partner. Project Glasswing is described by the company as “an urgent project to secure the world’s most critical software.” Paraphrast.

Anthropic Discussion

• Anthropic’s Claude Mythos Preview is scary!
• Anthropic will offer twice as much Claude for the following two weeks. I am now a good time to jump ship from ChatGPT/Gemini for smuggler.
• Chinese AI developers are copying AI models?

Explore More…

The Claude Mythos Preview is an unreleased model exclusive to the abovementioned partners at its core, and is known as “the Jacques Mythhos” (as it has been called) models. Furthermore, Anthropic also provides access to a group of over 40 organisations that develop or maintain critical software infrastructure. It has also invested up to $100 million (about Rs. 923 Paraphrasing of. In usage credits, 6 crore) was used and $4 million (roughly Rs. DIRECT donations of 37 crore) to open-source security organisations were made.

Anthropic says the AI model, which is based on cybersecurity, has “already found thousands of high-severity vulnerabilities” including some in every major operating system and web browser. The term ’I understand and modify complex software to identify security defects, as well with repair them. In their defense security work, the model will be deployed by the partnering organisations as part of its deployment of .

Although Claude Mythos Preview is a gated research preview, it’s largely paid model; $25 per million input and $125 permillion output tokens are for the pay model. This allows the partners to use it via the Claude application programming interface (API), Amazon Bedrock, Google Cloud’s Vertex AI and Microsoft Foundry.

“We do not plan to make Mythos Preview available in general. In a post, the company said ‘We want to deploy Mythos-class models safely at scale.’ First we need safeguards that reliably block their most dangerous outputs.