Anthropic Opens Access to Its Most Powerful AI Model to Security Teams, Discloses 129,000 Vulnerabilities

Anthropic has officially opened access to its most advanced AI model to more cybersecurity teams through an updated Cyber Verification Program (CVP). The move follows the Project Glasswing initiative, which identified more than 129,000 verified software vulnerabilities between April and July this year. The company said the figure is likely far below reality, estimating the true impact could be five times higher because the new data came from only a subset of partners.

The new program merges two earlier initiatives: Glasswing, which gave access to the Claude Mythos model family to organizations securing critical software, and the initial CVP, which gave verified security teams reduced restrictions on the Claude Opus and Sonnet models. Now, CVP comes with three tiers—Defense, Red Team, and Specialized—each with different verification requirements and security controls. All tiers get access to Claude Opus 5.5, Sonnet 5.5, Mythos 5.1, and upcoming models.

The launch of Claude Mythos Preview in April had raised concerns that AI could be exploited to hack software before it is secured. However, Anthropic appears to want to flip that narrative by making AI a defense tool. The Defense tier covers incident response and malware analysis, open to security teams, critical infrastructure operators, open-source maintainers, and researchers with a track record of vulnerability reporting. The Red Team tier adds penetration testing and red-teaming, but only organizations can apply. Meanwhile, the Specialized tier, with the loosest restrictions, is only for a small group of organizations authorized to test critical safety systems such as power grids, aviation systems, and interbank transfer infrastructure.

"We verify every member together with the US government, and existing Glasswing members will be moved to this tier," Anthropic said in its announcement.

For Indonesia, this development offers a valuable lesson. Amid accelerating digital transformation, the country still faces a shortage of cybersecurity talent and limited access to cutting-edge AI technology. If models like Claude can help identify security gaps on a massive scale, Indonesia could potentially use them to protect vital infrastructure—from banking to government systems. However, dependence on foreign companies and US government oversight raises questions about data sovereignty and geopolitical risk. Domestic regulators need to encourage strategic partnerships with global AI providers, while strengthening the national cybersecurity regulatory framework.

Going forward, Anthropic's business model could become a new industry standard: AI not only as an offensive tool, but also as a defensive goalkeeper. The question is whether developing countries like Indonesia will be granted equal access, or instead be left behind in a cyber arms race increasingly determined by artificial intelligence?