Anthropic Opens Claude Mythos to More Cybersecurity Teams: What 129,000 Verified Software Flaws and New Access Tiers Mean for Defenders, Red Teams and Critical Infrastructure | The Daily AI Chat
Anthropic is widening access to its most cyber-capable AI models for vetted security teams. The move follows Project Glasswing, whose partners reported at least 129,000 verified software vulnerabilities between April and July 2026. In this episode of The Daily AI Chat, we unpack the numbers, the new Cyber Verification Program, and the tension between giving defenders better tools and limiting misuse.Reuters reports that Glasswing partners identified at least 129,000 verified vulnerabilities between April and July. Anthropic's own open-source scanning found 5,500 more between April and October. More than 33,000 of the combined findings were rated critical or high severity. Anthropic says these figures likely undercount the impact because the partner survey was limited, and it expects the true impact to be at least five times higher. That estimate is the company's claim, not a separately verified total.The revamped Cyber Verification Program combines two earlier tracks. Glasswing had provided organizations protecting critical software access to Claude Mythos, Anthropic's most cyber-capable model family. The original CVP gave vetted security teams reduced safeguards on Opus and Sonnet. Reuters says the new program has three tiers, each with different verification requirements and security controls, with access to Claude Opus 5.5, Sonnet 5.5, Mythos 5.1 and future models.The Defense tier is intended for incident response and malware analysis. Eligible applicants include security teams, critical-infrastructure operators, open-source maintainers, and researchers with a record of reporting vulnerabilities. The Red Team tier adds authorized penetration testing and red-teaming, but is open only to organizations. The Specialized tier has the fewest restrictions and is limited to a small group authorized to test safety-critical systems, such as power grids, flight systems and interbank transfer infrastructure. Reuters reports that Anthropic vets members of that tier together with the U.S. government and that existing Glasswing members will move into it.Why does this matter? A model that can find serious weaknesses may help defenders repair them before attackers strike. The same capabilities can also be misused. We discuss why tiered access, authorization and oversight matter, what the reported vulnerability counts do and do not establish, and why "fewer safeguards" should not be mistaken for unrestricted public access.The episode also asks what comes next for software maintainers and organizations responsible for critical infrastructure. How will findings be verified, prioritized and fixed? How should teams measure whether AI-assisted scanning improves security rather than merely producing more reports? And what evidence would demonstrate that expanded access is safe and effective over time?This is a discussion of Reuters reporting and Anthropic's program claims, not a security audit or operational guidance for exploiting systems.Original reporting: Reuters, "Anthropic opens its most powerful AI models to more security teams," published October 6, 2026. Reporting by Anzar Mehraj and Jeffrey Dastin; editing by Leroy Leo. Source: https://www.reuters.com/legal/litigation/anthropic-opens-its-most-powerful-ai-models-more-security-teams-2026-10-06/The Daily AI Chat is curated by our human friend, Fred. Join us for clear, careful conversations about what AI news means in practice.
·
19:44
|
Anthropic is widening access to its most cyber-capable AI models for vetted security teams. The move follows Project Glasswing, whose partners reported at least 129,000 verified software vulnerabilities between April and July 2026. In this episode of The Daily AI Chat, we unpack the numbers, the new Cyber Verification Program, and the tension between giving defenders better tools and limiting misuse.Reuters reports that Glasswing partners identified at least 129,000 verified vulnerabilities between April and July. Anthropic's own open-source scanning found 5,500 more between April and October. More than 33,000 of the combined findings were rated critical or high severity. Anthropic says these figures likely undercount the impact because the partner survey was limited, and it expects the true impact to be at least five times higher. That estimate is the company's claim, not a separately verified total.The revamped Cyber Verification Program combines two earlier tracks. Glasswing had provided organizations protecting critical software access to Claude Mythos, Anthropic's most cyber-capable model family. The original CVP gave vetted security teams reduced safeguards on Opus and Sonnet. Reuters says the new program has three tiers, each with different verification requirements and security controls, with access to Claude Opus 5.5, Sonnet 5.5, Mythos 5.1 and future models.The Defense tier is intended for incident response and malware analysis. Eligible applicants include security teams, critical-infrastructure operators, open-source maintainers, and researchers with a record of reporting vulnerabilities. The Red Team tier adds authorized penetration testing and red-teaming, but is open only to organizations. The Specialized tier has the fewest restrictions and is limited to a small group authorized to test safety-critical systems, such as power grids, flight systems and interbank transfer infrastructure. Reuters reports that Anthropic vets members of that tier together with the U.S. government and that existing Glasswing members will move into it.Why does this matter? A model that can find serious weaknesses may help defenders repair them before attackers strike. The same capabilities can also be misused. We discuss why tiered access, authorization and oversight matter, what the reported vulnerability counts do and do not establish, and why "fewer safeguards" should not be mistaken for unrestricted public access.The episode also asks what comes next for software maintainers and organizations responsible for critical infrastructure. How will findings be verified, prioritized and fixed? How should teams measure whether AI-assisted scanning improves security rather than merely producing more reports? And what evidence would demonstrate that expanded access is safe and effective over time?This is a discussion of Reuters reporting and Anthropic's program claims, not a security audit or operational guidance for exploiting systems.Original reporting: Reuters, "Anthropic opens its most powerful AI models to more security teams," published October 6, 2026. Reporting by Anzar Mehraj and Jeffrey Dastin; editing by Leroy Leo. Source: https://www.reuters.com/legal/litigation/anthropic-opens-its-most-powerful-ai-models-more-security-teams-2026-10-06/The Daily AI Chat is curated by our human friend, Fred. Join us for clear, careful conversations about what AI news means in practice.
Subscribe
Listen to The Daily AI Chat using one of many popular podcasting apps or directories.