Anthropic Launches Project Glasswing to Defend Against AI-Powered Cyberattacks

Anthropichas announced Project Glasswing, an initiative designed to secure critical software infrastructure against emerging AI-powered threats. The program enlists 11 major technology partners: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks.

Participants will use Claude Mythos Preview, an unreleased general-purpose AI model from Anthropic, to enhance their security operations. According to Anthropic, this model has identified thousands of exploitable vulnerabilities, including in every major operating system and web browser.

Model Details and Scope

Clause Mythos Preview represents Anthropic's approach to defensive cybersecurity applications. The company states the initiative aims to prevent malicious AI deployment that could create severe economic and security consequences. Rather than releasing the model publicly, Anthropic is restricting access to vetted security organizations and critical infrastructure partners.

The breadth of the partner consortium—spanning cloud providers, hardware manufacturers, security firms, and financial institutions—suggests the initiative targets vulnerabilities across multiple attack surfaces and technology stacks.

Context and Previous Actions

This announcement reflects Anthropic's stated commitment to AI ethics and responsible deployment. Earlier in 2025, Anthropic refused to remove safety guardrails on its Claude service for U.S. Department of Defense applications, which resulted in the DoD designating Anthropic a "supply chain risk." The company has maintained strict policies around its models' use cases despite pressure from government and military entities.

However, Anthropic's own Claude model was reportedly leveraged by a hacker in February to attack multiple government agencies in Mexico, underscoring the dual-use vulnerability inherent in capable AI systems.

What This Means

Project Glasswing positions Anthropic as a proactive actor in AI safety infrastructure rather than solely a model provider. By restricting access to Claude Mythos Preview to security-focused partners, Anthropic attempts to capture the defensive applications of advanced AI while limiting offensive exploitation risk. The partnership roster indicates serious commitment from major infrastructure stakeholders, though the initiative's real-world impact will depend on execution and whether identified vulnerabilities are actually remediated at scale. The timing also serves Anthropic's strategic interests following the DoD designation, demonstrating tangible national security contributions outside military applications.