Anthropic Expanded Its Cyber Verification Program
The initiative grants vetted security teams access to modified AI models to identify software vulnerabilities.
Updated on Oct. 6, 2026 in Cybersecurity
Live Poll
Should tech companies provide security teams broader access to powerful AI models for software testing?
Anthropic has launched an expanded Cyber Verification Program that provides security teams access to Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1 models with reduced safety safeguards. This effort formalizes and combines two internal initiatives that operated over the last six months.
Why it matters
The program aims to secure critical global software infrastructure by leveraging AI to identify vulnerabilities at scale. Anthropic conducts member vetting in coordination with the United States government to mitigate risks associated with providing access to less-restricted models.
The program has identified over 100,000 software vulnerabilities this year, with at least 33,000 classified as critical or high severity. The initiative operates across three distinct tiers: Defense, Red Team, and Specialized.
The players
Anthropic
An AI research and deployment company focused on building large language models with specific safety architectures.
The details
The program grants vetted members access to specific AI models, including the Claude Opus 5.5 and Mythos 5.1, which function with fewer safety constraints than standard consumer-facing versions. These professionals use the models to perform automated security analysis and code review to flag potential exploits. The three-tier structure allows Anthropic to apply tiered verification requirements and security controls based on the level of model access required for each task.
Timeline
April 2026 marked the start of the vulnerability scanning period.
Data collection for verified vulnerabilities concluded in July 2026.
Anthropic announced the revamped program on October 6, 2026.
The Tech Race
This program integrates the previous findings of Project Glasswing into a more structured, multi-tier deployment framework. It signals a shift toward formalized, government-coordinated AI red teaming to address the massive volume of high-severity software vulnerabilities.
The program changes workflows for professional security teams by providing access to more capable, less-restricted AI models. These tools are currently restricted to vetted members and are not available for public use.
The takeaway
Anthropic is moving to scale automated vulnerability detection by giving security researchers more permissive model access. The key metric to watch is whether the program's verified vulnerability counts scale in line with the company's projection of five-times higher impact.
Further reading
For more on how AI is being deployed in threat detection, see our Cybersecurity section.
Live Poll
Should tech companies provide security teams broader access to powerful AI models for software testing?
