Anthropic opens cyber program and launches Claude Haiku 5.5

Anthropic has expanded its Cyber Verification Program (CVP) to give qualified security professionals better access to defensive AI tools. Originally introduced in April alongside Project Glasswing and Claude Mythos, the updated program features three access tiers and fewer blocking classifiers. Approved participants can access powerful models including the previously restricted Claude Mythos 5.1, the high-end Claude Opus 5.5, and the balanced Claude Sonnet 5.5. While the program itself is free, accepted users must pay standard token-consumption rates and prove they possess rigorous security controls to prevent model leaks.

This expansion comes as Western AI developers face intense pressure from Chinese open-weight models, such as Z.ai's GLM-5.2 and GLM-5.3, which provide capable threat detection at a lower cost. Security experts note that widespread access to slightly less capable open models can sometimes be more dangerous than restricted access to top-tier proprietary systems. To counter this, Anthropic is positioning its closed ecosystem as a secure, trusted alternative, competing alongside rival initiatives like OpenAI's Daybreak, Google's Fairwind, and the CrowdStrike-Nvidia SafeMind suite built on Nemotron models.

Alongside the CVP expansion, Anthropic released Claude Haiku 5.5, a model optimized for speed-sensitive applications like customer support and browser use. The company has structured its pricing to charge $0.10 per input token up to 100,000 tokens, rising to $0.50 per input token for volumes over 100,000. For output, the model costs $0.50 per token up to 100,000 tokens and $2.50 per token above that threshold. This pricing structure contrasts with OpenAI's GPT-6 Luna, which is priced at $0.10 per million input tokens and $0.50 per million output tokens.

For enterprise security practitioners, these updates offer a dual advantage. They gain access to highly sophisticated, vetted models for vulnerability discovery and offensive testing without the risk of open-source leakage. At the same time, the introduction of Claude Haiku 5.5 provides a highly economical option for high-volume, low-latency tasks, allowing developers to balance heavy-duty security analysis with cost-effective daily operations.