Anthropic's Selective Access to AI Models

Anthropic’s recent product announcements have clarified its approach to cybersecurity partnerships. At the forefront are major security firms like CrowdStrike, Palo Alto Networks, and Zscaler, which have been granted access to the advanced Claude Mythos Preview AI model through Project Glasswing. These companies are also embedding Anthropic's prominent model, Claude Opus 4.7, into their security systems.

In contrast, vendors such as SentinelOne and TrendAI have managed to integrate only Opus 4.7 into their offerings. However, they do not have access to Mythos Preview, which Anthropic has decided to restrict due to concerns about its potential misuse by adversaries.

Anthropic’s tiered access strategy stands in stark contrast to OpenAI’s more inclusive method for GPT-5.4-Cyber, where multiple security firms, including Cloudflare and iVerify, received access based on clear criteria like identity verification. This difference highlights the elevated risk associated with Mythos as a new class of AI model.

The financial strength of the companies involved also varies significantly. CrowdStrike and Palo Alto Networks are valued at $119 billion and $150 billion, respectively, with substantial annual revenues. On the other hand, SentinelOne and Trend Micro are valued at approximately $5 billion, reflecting their lesser capital resources compared to top-tier partners.

Opus 4.7 allows partners to identify complex issues like logic flaws and supply-chain risks across entire systems, surfacing vulnerabilities that often escape detection through traditional methods. Meanwhile, Mythos Preview provides superior insights—Palo Alto can identify vulnerabilities that prior models missed entirely.

As Anthropic continues to forge partnerships, it is also collaborating with leading firms like Accenture and Deloitte to implement Claude-integrated solutions. These initiatives are indicative of the industry's shift toward AI-driven security frameworks.

As Anthropic's market share grows—from 12% in 2023 to an expected 32% by 2025—its selective access model emphasizes the importance of robust partnerships in navigating today’s cybersecurity landscape.