How the Tiered System Works
Anthropic, the maker of Claude, has broadened its Cyber Verification Program (CVP) this week, introducing a tiered access system based on use cases and adding its Project Glasswing to the most open tier. The change is aimed at expanding the reach of the tool to more cyber defenders, though the exact number of new users remains unclear.
Breaking news
Town AI Secures $108 Million Series B, Valued at $1 Billion After $55 M Pre‑Series
Apple Prepares for Major Product Launch Event on October 28
Apple Releases Release Candidates for watchOS, tvOS and visionOS 27.2
Apple Limits iPhone Duo and iPhone 18 Pro Max to Reservation-Only Sales in Japan and South KoreaThe new structure divides CVP into three tiers: a highly restricted level for sensitive applications, a moderate tier for general security work, and an open tier that now includes Project Glasswing. Project Glasswing, a platform that helps organizations test and validate AI models for safety, was previously limited to a smaller group of partners. By moving it to the open tier, Anthropic claims it will be available to a broader set of defenders who need rapid, reliable AI safety checks.
Will the Expansion Reach All Defenders? (Question)
Anthropic’s announcement came during a briefing in San Francisco, where the company highlighted that the tiered approach allows for more nuanced control over how the model is used. The firm stated that the new framework will „extend the impact of our Project Glasswing to a much larger number of cyber defenders.” The move is part of a broader trend in AI safety, where companies are balancing openness with risk mitigation. Critics note that while the open tier offers greater accessibility, the restricted tiers still limit many organizations that could benefit from AI‑driven security tools.
The tiered system assigns access based on the sensitivity of the user’s application. The highest tier is reserved for organizations dealing with classified or highly confidential data, requiring strict compliance checks. The middle tier covers typical enterprise security teams, allowing them to run standard threat‑analysis tasks. The lowest tier, now inclusive of Project Glasswing, offers the most flexibility, letting smaller firms and independent researchers experiment with the model without extensive vetting.
Consequences and Outlook
According to Anthropic’s spokesperson, the tiered approach was designed to reduce the risk of misuse while still encouraging innovation. „We wanted to create a pathway that lets more people benefit from our safety tools without compromising the integrity of the system,” the spokesperson said. The company also noted that the open tier will include enhanced monitoring to detect potential abuse early.
While the open tier promises broader access, the exact number of new users is not yet disclosed. Some cybersecurity experts worry that the tiered model may still exclude smaller organizations that lack the resources to navigate the application process. Others argue that the added oversight could slow deployment in urgent threat environments. The balance between accessibility and safety remains a key debate in the AI security community.
The announcement also hinted at future plans to integrate additional safety features into the open tier, such as real‑time monitoring dashboards and automated compliance reporting. These enhancements aim to make the platform more user‑friendly for non‑technical defenders.
Frequently Asked Questions
The expanded CVP could accelerate the adoption of AI in cybersecurity, allowing more defenders to test and validate models quickly. However, the tiered system may also create a fragmented user base, with some organizations gaining full access while others remain restricted. Over time, Anthropic may need to adjust the thresholds to keep pace with evolving threats and user needs.
Ultimately, the move signals a shift toward more flexible AI safety frameworks. If successful, it could set a precedent for other AI developers to adopt tiered access models that balance openness with risk mitigation.
