Claude Mythos Found 231 Microsoft Bugs in a Month — And Microsoft Can't Patch Fast Enough

In April 2026, Anthropic's Claude Mythos Preview AI model identified 231 vulnerabilities in Microsoft products — a volume that has overwhelmed Microsoft's patching capacity, according to internal Microsoft recordings and documents reported by ProPublica and Ars Technica on July 29, 2026 . This single-month discovery set off a chain reaction: a record-breaking Patch Tuesday, a rare warning from the Five Eyes intelligence alliance, and urgent questions about whether the cybersecurity industry can keep up with AI-speed vulnerability discovery.

The Scale of Discovery: 231 Bugs in One Month

Internal Microsoft presentations show that in April 2026, Mythos uncovered 90 critical bugs and 141 important ones in SharePoint alone, with even higher volumes in the first half of May . The model, part of Anthropic's Project Glasswing initiative, is so effective that Microsoft has been forced to focus its engineering resources on only the most severe vulnerabilities, leaving many discovered flaws unpatched indefinitely .

More broadly, across Project Glasswing, Mythos has identified more than 10,000 high- or critical-severity vulnerabilities across the world's most systemically important software since the program launched . Anthropic's own coordinated vulnerability disclosure dashboard reports 1,596 disclosed vulnerabilities across 281 open-source projects as of May 22, 2026, with roughly 23,019 findings triaged and 1,900 candidates prioritized .

Microsoft's Internal Triage Crisis

The gap between discovery and remediation has been described as an existential triage crisis — the AI finds flaws faster than human patch teams can analyze, reproduce, and fix them . Internal Microsoft presentations reportedly acknowledge the company cannot keep pace with the Mythos-driven discovery rate . An internal slide from a Microsoft presentation noted that the company plans to eventually address moderate-severity flaws, but the immediate focus is on critical and important vulnerabilities only .

The Record-Breaking July 2026 Patch Tuesday

On July 14, 2026, Microsoft shipped its largest Patch Tuesday in history, addressing 622 vulnerabilities — triple the June 2026 count and roughly five times the monthly average . Counts vary slightly by source, with some reporting 569–622 CVEs . The release included:

• 59 critical-rated flaws
• Three zero-days, two of which — CVE-2026-56155 in Active Directory Federation Services and CVE-2026-56164 in SharePoint Server — were already being actively exploited before patches existed
• 416 Windows defects, plus fixes across Office, SharePoint, Azure, Hyper-V, and Visual Studio

CrowdStrike noted this release "reinforces that the era of 'small' Patch Tuesdays may be over as AI-driven vulnerability discovery ramps up" . Just one month earlier, June 2026 had already broken the previous record with 206 CVEs .

Hitachi Joins Project Glasswing for Critical Infrastructure

Hitachi joined Project Glasswing in June 2026, gaining access to Claude Mythos Preview to audit its social infrastructure software — particularly in the energy sector . On July 28, 2026, Hitachi reported that Mythos reduced security validation processes from weeks to dramatically shorter timeframes, successfully creating threat models for millions of lines of source code and confirming the model's practical applicability to critical infrastructure .

Five Eyes Intelligence Warning: 'The Timeline Is Not Years, It Is Months'

On June 22, 2026, the Five Eyes intelligence alliance (US, UK, Canada, Australia, New Zealand) issued a rare joint public statement warning that cutting-edge AI models are poised to supercharge offensive hacking capabilities within months, not years . The alliance stated: "Frontier AI models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities. The timeline is not years, it is months" . The warning came weeks after the Trump administration ordered Anthropic to restrict access to its Mythos and Fable models for foreign nationals . While the Five Eyes statement did not name specific companies, Reuters and other outlets linked the urgency directly to concerns about Anthropic's Mythos model .

Broader Implications: The Patch Velocity Mismatch

The Mythos discovery rate has exposed a fundamental mismatch between AI-powered vulnerability discovery and human-led remediation. Security experts have raised concerns:

• Patch velocity mismatch: AI discovers flaws at machine speed, but human triage and patch development remain constrained by manual processes .
• Backlog risk: Thousands of high-severity flaws stacking up force impossible prioritization decisions, leaving many systems exposed .
• Infrastructure fragility: Hitachi's work shows AI-driven discovery now targets industrial control systems and energy infrastructure, where patching is even slower due to uptime requirements and regulatory constraints .
• Offensive asymmetry: The same AI capability used defensively in Project Glasswing could be weaponized by adversaries, creating an unprecedented asymmetry in cyber operations .

As one security analyst noted: "Before, we worried about finding bugs. Now, we worry about finding them faster than we can fix them. The bottleneck has shifted from discovery to remediation — and we don't have AI for that yet."

What Organizations Should Do Now

The Five Eyes advisory listed five practical actions for businesses: reduce attack surface, patch more quickly, assume breach, invest in detection and response, and harden identity infrastructure . For most enterprise security teams, this means rethinking vulnerability management programs that were designed for a human-paced discovery world. Automated patching, AI-assisted triage, and risk-based prioritization are no longer optional — they are survival tactics.

Anthropic's Claude Mythos has proven that AI can find vulnerabilities at unprecedented scale and speed. The question now is whether the global cybersecurity ecosystem can evolve fast enough to patch them.