Anthropic, on Tuesday, introduced Undertaking Glasswing, a brand new initiative centered on synthetic intelligence (AI)-powered cybersecurity. Powering the undertaking for the San Francisco-based AI agency is a general-purpose mannequin dubbed Claude Mythos Preview. The corporate introduces the mannequin as its most succesful in relation to coding and agentic duties. For now, it’s only obtainable to the companions of Undertaking Glasswing and can allow them to detect safety flaws and vulnerabilities throughout their tech stack. The AI agency has additionally allotted monetary sources for the initiative.
Anthropic’s Undertaking Glasswing is Right here
In a put up on X (previously referred to as Twitter), the AI startup introduced the initiative, together with launch companions Amazon Internet Companies (AWS), Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Basis, Microsoft, Nvidia, and Palo Alto. The corporate describes Undertaking Glasswing as “an pressing initiative to assist safe the world’s most crucial software program.”
At its core is the Claude Mythos Preview, an unreleased mannequin unique to the abovementioned companions. Moreover, Anthropic can also be extending entry to a bunch of greater than 40 organisations that construct or keep essential software program infrastructure. The corporate has additionally dedicated as much as $100 million (roughly Rs. 923.6 crore) in utilization credit, in addition to $4 million (roughly Rs. 37 crore) in direct donations to open-source safety organisations.
Anthropic claims that the cybersecurity-focused AI mannequin has “already discovered 1000’s of high-severity vulnerabilities, together with some in each main working system and net browser.” It’s stated to grasp and modify complicated software program to search out safety flaws, in addition to repair them. The partnering organisations will deploy the mannequin as a part of their defensive safety work.
Regardless of being a gated analysis preview, Claude Mythos Preview is a paid mannequin, priced at $25 per million enter and $125 per million output tokens. It’s accessible to the companions by way of the Claude software programming interface (API), Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.
“We don’t plan to make Mythos Preview typically obtainable. Our purpose is to deploy Mythos-class fashions safely at scale, however first we’d like safeguards that reliably block their most harmful outputs,” the corporate stated in a put up.
