Anthropic Expands AI Access For Cyber Defenders
Anthropic is expanding access to its most powerful artificial intelligence models for cybersecurity professionals, allowing vetted security teams to use advanced capabilities with fewer safety restrictions.
The company has revamped its Cyber Verification Program (CVP), combining it with its earlier Project Glasswing initiative. The aim is to give legitimate cybersecurity teams more powerful AI tools to identify, analyse and fix vulnerabilities before attackers can exploit them.
Three Levels Of Access Introduced
The expanded programme now has three access tiers, each designed for different types of cybersecurity work.
The Defense tier is the broadest and is intended for activities such as incident response, malware analysis, vulnerability research and security operations.
The Red Team tier provides additional capabilities for authorised penetration testing and adversarial security assessments. Only organisations can apply for this level.
The Specialized tier has the fewest restrictions and is reserved for a small number of highly vetted organisations testing safety-critical infrastructure.
This can include systems supporting power grids, aviation, telecommunications, financial networks and government operations.
Claude Opus 5.5 And Mythos 5.1 Included
All three tiers provide access to Anthropic’s advanced models, including Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, along with future models.
The company is reducing certain safety blocks for approved users because conventional AI safeguards can sometimes prevent legitimate security researchers from carrying out defensive work.
However, Anthropic says dangerous activities such as deploying ransomware or causing physical damage will continue to face restrictions in applicable tiers.
AI Has Already Found Thousands Of Vulnerabilities
The expansion follows Anthropic’s Project Glasswing programme, under which partner organisations used AI to search for weaknesses in important software.
According to Anthropic, its partners identified at least 129,000 verified software vulnerabilities between April and July 2026.
More than 33,000 of those vulnerabilities have so far been classified as critical or high severity.
Anthropic’s own open-source scanning efforts identified another 5,500 verified vulnerabilities between April and October.
The company cautions that these figures are likely an undercount.
The Double-Edged Sword Of AI Cybersecurity
The decision highlights a growing challenge for AI companies.
The same technology that can help security researchers discover vulnerabilities can also help attackers exploit them.
Anthropic therefore requires applicants to undergo verification and demonstrate appropriate security controls. The most sensitive access level involves detailed reviews conducted in collaboration with the US government.
The company says the goal is to give defenders access to advanced capabilities while preventing those capabilities from being easily misused.
AI Could Change Cybersecurity
The expansion signals a major shift in how frontier AI models may be used for cybersecurity.
Instead of keeping the most capable systems behind broad restrictions, Anthropic is creating controlled environments where trusted security professionals can use them for legitimate defensive work.
If AI continues finding vulnerabilities at the scale demonstrated through Project Glasswing, such systems could become an increasingly important part of vulnerability research, incident response and critical infrastructure protection.
Summary
Anthropic has expanded its Cyber Verification Program to give vetted cybersecurity professionals greater access to its most powerful AI models with fewer safety restrictions. The programme now has Defense, Red Team and Specialized tiers. Anthropic says Project Glasswing partners discovered at least 129,000 verified vulnerabilities between April and July, including more than 33,000 rated critical or high severity.
