--- url: "https://www.xcademia.com/news/anthropic-expands-cyber-verification-program-with-three-access-tiers" title: Anthropic Expands Cyber Verification Program With Three Access Tiers description: Anthropic expands its Cyber Verification Program with three access tiers for verified security teams and different levels of cyber safeguards. publishedAt: "2026-10-07T07:06:52.102+00:00" updatedAt: "2026-10-10T05:47:26.773424+00:00" type: news category: cybersecurity source_name: Anthropic source_url: "https://www.anthropic.com/news/cyber-verification-program" tags: - "#Cybersecurity" - "#AI" - "#Anthropic" - "#AISecurity" - "#CyberDefense" - "#VulnerabilityResearch" - "#RedTeam" - "#CloudSecurity" --- # Anthropic Expands Cyber Verification Program With Three Access Tiers > Anthropic is expanding its Cyber Verification Program with three access tiers for security professionals, offering different levels of cyber capabilities and safeguards based on the type and risk of authorized security work. Source: **Anthropic** · 7 October 2026 ## Anthropic Expands Cyber Access for Verified Security Teams Anthropic is expanding its Cyber Verification Program (CVP), introducing three access tiers designed to provide verified security professionals with different levels of access to advanced cyber capabilities and reduced blocking classifiers. The company says the expanded program combines capabilities previously offered through its Cyber Verification Program and Project Glasswing. Each tier includes access to Anthropic's most capable models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward. The program reflects the dual-use nature of cybersecurity. The same AI capabilities that can help defenders identify vulnerabilities, analyze malware, and investigate incidents can also be used for harmful activity. Anthropic says the expanded CVP is intended to provide broader access for legitimate security work while maintaining additional restrictions around higher-risk activities. ## Three Access Levels for Cybersecurity Work The three CVP tiers organize access according to the type of security work and the level of authorization involved. ### (1) Defense Access Defense Access is designed for organizations and researchers conducting defensive cybersecurity work. This includes activities such as: - Security operations center and incident response work - Malware reverse engineering - Vulnerability analysis and validation - Security research on systems owned or maintained by the organization Anthropic says qualifying users can include security teams at companies, nonprofits, universities, and government organizations. The company also identifies operators of critical infrastructure, such as regional hospitals and municipal utilities, as potential qualifying organizations. Smaller security firms, open-source maintainers, and individual researchers with a track record of reporting vulnerabilities can also qualify. Anthropic expects many defensive organizations to qualify and says it aims to respond to applications within a few days. ### (2) Red Team Access Red Team Access adds authorized penetration testing and red-team activities to the defensive uses covered by Defense Access. This tier is intended for organizations such as: - In-house red teams - Government red teams - Security firms - Penetration testing organizations Testing must be authorized and conducted against systems the organization has permission to test, including IT systems supporting critical industries. Anthropic says some restrictions remain in place. Real-time blocks continue to apply to actions that could cause physical harm or mass disruption, including deploying ransomware, damaging physical systems, or conducting penetration testing against high-risk safety systems. Red Team Access is available only to organizations, not individual researchers. Anthropic expects applications for this tier to take a few weeks to review. Applicants are enrolled in Defense Access while their Red Team application is being evaluated. ### (3) Specialized Access Specialized Access provides the fewest cyber-related blocks and is limited to a small number of verified organizations. The tier is intended for organizations authorized to test systems where failures could impact lives or disrupt markets. Anthropic gives examples including: - Flight operating systems - Power grids - Telecommunications networks - Interbank transfer infrastructure - Government administrative networks Organizations applying for Specialized Access undergo an in-depth review conducted in collaboration with the US government. Existing Project Glasswing members transition to Specialized Access without requiring another approval for current models. ### ![info-1](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1791356493219-info-1--198-.webp) ## General-Purpose Models Continue to Use Conservative Cyber Safeguards The expanded CVP does not remove cybersecurity safeguards from Anthropic's generally available models. Anthropic says generally available models such as Claude Opus 5.5, Claude Fable 5.1, and Claude Sonnet 5.5 continue to use conservative cyber safeguards. These safeguards are designed to block most cyber-related work while reducing false positives for legitimate secure coding activities. Users can still use generally available models for activities such as: - Code review - Patching known issues - Finding vulnerabilities in source code they own - Security alert triage The expanded CVP provides additional access for verified professionals whose work requires capabilities beyond those generally available safeguards. ## Anthropic Gives CVP Participants Access to Advanced Models Anthropic says CVP participants receive access to advanced models based on their approved tier. The company specifically lists Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1, along with new models moving forward. The program brings together capabilities previously provided through separate initiatives. Over the previous six months, Project Glasswing provided a group of organizations securing critical software with access to Claude Mythos. The existing CVP also provided vetted security teams with reduced safeguards on Claude Opus and Claude Sonnet. Anthropic has now integrated these efforts into the expanded program. ## CyScenarioBench Tests Different Safeguard Levels Anthropic also published results from an evaluation called CyScenarioBench. The evaluation examines an AI model's ability to plan and execute multi-stage cyber operations under realistic constraints. For the evaluation, Anthropic tested Claude Opus 5.5 across 10 challenges, with five attempts for each tested access configuration, producing 50 trials per configuration. The results were: Configuration Result Without CVP Every task was blocked on the first prompt Defense Access 46 of 50 trials were blocked at some point; 4 completed Red Team Access No blocks; 34 of 50 trials completed No safeguards comparison 67.6% success rate Under Defense Access, 46 of the 50 trials were blocked at some point, while the remaining four trials completed. Under Red Team Access, Anthropic reported no blocks and 34 completed trials out of 50. Anthropic says the 34-of-50 result is effectively equivalent to a 67.6% success rate when no safeguards were applied, which it describes as representative of Specialized Access. These results are Anthropic's own evaluation results and should not be interpreted as an independent assessment of the Cyber Verification Program. ### ![info-2](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1791356509334-info-2--180-.webp) ## Project Glasswing Found More Than 129,000 Verified Vulnerabilities Anthropic also provided new information about Project Glasswing, its initiative involving organizations working on cybersecurity and critical software. According to Anthropic, Project Glasswing partners uncovered at least **129,000 verified software vulnerabilities** between April and July 2026. Anthropic's own open-source scanning identified another **5,500 verified vulnerabilities** between April and October 2026. More than **33,000** of the verified vulnerabilities were rated critical or high. Anthropic cautions that these figures are likely an undercount because the data comes from only a subset of Project Glasswing partners. The company estimates that the true impact could be at least five times higher. This is an Anthropic estimate and was not independently verified in the announcement. Several partners also told Anthropic that AI models increased their vulnerability discovery rate. According to the company, some partners said finding the same number of vulnerabilities without Claude Mythos could have taken months or years. Anthropic cited Booz Allen and Comcast among the partners involved. These statements represent partner feedback reported by Anthropic and are not independent measurements of time savings. ### ![info-3](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1791356523862-info-3--168-.webp) ## Patch Data Remains Incomplete The vulnerability figures do not fully reflect remediation. Anthropic says fewer than half of the participating partners disclosed the number of vulnerabilities that had been patched. In many cases, fixes were still in progress. As a result, Anthropic says the reported patch rate significantly undercounts remediation progress. The announcement therefore provides substantially more information about vulnerability discovery than about completed remediation. The announcement did not disclose additional details. ## CVP Adds Verification and Data Controls The expanded program includes additional verification requirements for participating organizations. Applicants must be verified and provide proof of the required security controls. Anthropic also provides a process for reporting false positives and appealing blocking decisions. Data retention is required for CVP organizations so Anthropic can monitor potential cyber misuse. The company says its Enterprise Frontier Safeguards (EFS), expected later in the fall, will combine zero-data-retention privacy with robust safeguards. Eligible organizations will be able to store data in cloud infrastructure they control. Until EFS becomes available, organizations with access to Claude Fable 5.1 or Claude Mythos 5.1 with zero data retention can also use CVP with zero data retention. Additional details about EFS were not disclosed in the announcement. ## What the Expanded Program Means for Enterprise Security The announcement highlights a broader industry shift toward treating AI cybersecurity capabilities according to the context in which they are used. Instead of applying one uniform level of access, Anthropic is separating defensive research, authorized red teaming, and highly sensitive infrastructure testing into different verification tiers. For enterprises, this could mean AI security programs increasingly need stronger controls around authorization, user eligibility, infrastructure scope, and data handling. The approach also highlights an ongoing challenge for AI security providers: advanced cyber capabilities can be useful for legitimate defenders while creating additional risks when applied to unauthorized or high-impact systems. Anthropic's tiered model is one approach to managing that dual-use problem. ## Anthropic Extends CVP Across Its AI Platform Ecosystem The expanded CVP is available through several platforms. Anthropic lists: - Claude Platform - Google Cloud Vertex AI - Microsoft Foundry - Amazon Bedrock Amazon Bedrock access is limited to customers eligible for Enterprise Frontier Safeguards. Existing CVP members will retain their current settings for previous models. They will also be automatically evaluated for Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. Anthropic says administrators can assign approved access to specific workspaces. The company also says it will continue refining its tier-based classifiers and plans to share more about this work in the coming weeks. ## Conclusion Anthropic's expanded Cyber Verification Program introduces a three-tier model for providing advanced AI cybersecurity capabilities to verified security professionals. Defense Access focuses on defensive security work. Red Team Access adds authorized penetration testing and red teaming. Specialized Access is reserved for a limited group of organizations testing systems where failures could impact lives or disrupt markets. The program also combines verification requirements, cyber safeguards, and data controls with access to Anthropic's advanced models. The Project Glasswing results provide additional context for why AI-assisted vulnerability discovery is becoming an important part of the cybersecurity discussion. At the same time, Anthropic's own figures come with limitations around partial reporting, incomplete patch data, and estimates that have not been independently verified. The expanded CVP represents Anthropic's effort to balance broader access to advanced cyber capabilities with additional controls for higher-risk security work. ## Original source https://www.anthropic.com/news/cyber-verification-program ## Tags `#Cybersecurity` · `#AI` · `#Anthropic` · `#AISecurity` · `#CyberDefense` · `#VulnerabilityResearch` · `#RedTeam` · `#CloudSecurity` --- ## About this content This Markdown news article is the citation-grade twin of [Anthropic Expands Cyber Verification Program With Three Access Tiers](https://www.xcademia.com/news/anthropic-expands-cyber-verification-program-with-three-access-tiers). It is published by **Xcademia** (UK Companies House 12322710) and is available for AI search engines and large language models to index, summarise, and cite. When citing or quoting, please attribute *Xcademia* and link back to the source URL above. - Source: https://www.xcademia.com/news/anthropic-expands-cyber-verification-program-with-three-access-tiers - Publisher: Xcademia, https://www.xcademia.com - Catalogue index: https://www.xcademia.com/llms-full.txt