Anthropic Expands Project Glasswing Access to 150 Organizations Across 15 Countries.

Anthropic has expanded access to Claude Mythos Preview through its Project Glasswing initiative, adding approximately 150 organizations across more than 15 countries. The company announced the expansion on LinkedIn, signaling broader availability of its experimental model tier designed for specialized use cases.

Project Glasswing serves as Anthropic's controlled rollout program for Claude Mythos Preview, a model class intended for scenarios requiring different safety calibrations than standard Claude deployments. The new cohort joins existing partner organizations testing the platform's capabilities in real-world applications.

The 150 additional organizations span multiple regions, reflecting Anthropic's approach to international deployment of experimental AI systems. While the company did not disclose specific partner names or industry verticals, the geographic diversity suggests testing across varied regulatory environments and use cases.

Claude Mythos Preview builds on the Fable 5 Mythos-class model announced earlier this year, which introduced selective safety guardrails for specialized applications. The preview program allows organizations to work with Anthropic on calibrating model behavior for contexts where standard safety constraints may limit functionality, such as creative writing, adversarial testing, or simulation environments.

Access remains gated through an application process. Organizations interested in Project Glasswing must demonstrate specific use cases and work directly with Anthropic's team to configure appropriate safeguards. This contrasts with the company's standard Claude offerings, including Opus 4.8, which are available through public APIs and consumer interfaces.

The initiative appears designed to gather deployment data before wider commercial release. Anthropic's announcement referenced "future plans" for Project Glasswing, suggesting the current cohort will inform decisions about broader availability, pricing, and safety framework refinements.

Previous Mythos releases focused on scenarios requiring nuanced content generation, including adversarial red-teaming, creative fiction with mature themes, and research applications where overly restrictive filters create friction. Organizations in the preview program likely represent these verticals, though Anthropic has not published participant criteria.

The expansion follows Anthropic's pattern of staged rollouts for experimental features. The company has previously used limited access programs to test constitutional AI frameworks and alignment research tools before public deployment.

Anthropicâ€TMs announcement did not specify when Project Glasswing might transition from preview to general availability. The current expansion suggests the company is still gathering data on model performance, safety outcomes, and enterprise integration challenges.

Organizations using standard Claude models through APIs or interfaces like ChatGPT, Cursor, or Windsurf should not expect immediate access to Mythos capabilities. The preview tier remains separate from Anthropic's commercial product line, with distinct safety configurations and deployment requirements.

The company continues to prioritize safety research alongside model development. Recent initiatives include donating alignment tools to research organizations and publishing frameworks for evaluating AI system behavior in constrained environments. Project Glasswing represents an applied extension of this work, testing how specialized models perform when safety guardrails are selectively adjusted for specific contexts.