570 Flaws, Three Zero-Days, and the AI That Found Them

 On July 14, Microsoft released security updates addressing 570 vulnerabilities, nearly triple the previous record and more than any single Patch Tuesday in the company’s history. Almost 60 of those bugs were rated Critical. Three were zero-days, two already exploited in the wild. The reason for the surge is as significant as the patches themselves: AI is now finding vulnerabilities at machine scale, and the entire patch management model is being forced to adapt. 

The Scale 

The 570 figure is not a gradual increase. Microsoft’s previous record (June 2026) was roughly 200 flaws. This month nearly triples that number. To put it in perspective: if every day of July were a Patch Tuesday, each day would still need to ship 18 fixes to reach this total. 

Microsoft Executive Vice President Pavan Davuluri addressed this directly in a July 9 blog post, explaining that AI-powered scanning is now running across the Windows codebase using MDASH, the multi-model agentic scanning harness, which uses “multiple models including leading third-party AI vulnerability discovery models” to scan critical binaries, validate candidates through multi-model debate, and feed high-confidence findings to engineering teams. The result is a larger volume of vulnerabilities discovered, filed, and fixed in each release cycle. 

Davuluri was explicit that customers should expect this to continue: “As AI helps defenders discover more issues, customers will see a higher volume of security updates included in each security release.” 

The Zero-Days 

Three zero-day vulnerabilities were addressed this month, two of which are being actively exploited: 

• CVE-2026-56155 – Elevation of privilege in Active Directory Federation Services (ADFS). Exploited in the wild. An attacker who successfully exploited this bug could elevate their user rights on a Windows system. 
• CVE-2026-56164 – Elevation of privilege in Microsoft SharePoint. Also actively exploited. Microsoft originally assigned this bug an exploitability rating of “Exploitation Less Likely,” yet CISA added it to the Known Exploited Vulnerabilities catalogue on July 1, weeks before Patch Tuesday. This discrepancy is discussed further below. 
•  CVE-2026-50661 – Security feature bypass in Windows BitLocker. Publicly disclosed but not yet exploited. An attacker with physical access could potentially gain access to encrypted data. This CVE highlights the ongoing tension between full-disk encryption and forensic access; the bypass undermines the core promise of BitLocker for devices that fall into adversarial hands. 

The Copilot RCE That Stands Out 

Among the Critical-rated vulnerabilities, CVE-2026-48561 deserves particular attention. This is a remote code execution flaw in Microsoft Copilot carrying a CVSS score of 9.6, near-maximum severity. The bug allows an unauthorized attacker to execute code over the network. The attack vector: an attacker hosts a malicious website that causes Microsoft Edge for Android to automatically send crafted prompts to Copilot when a user visits the site. 

This is notable for two reasons. First, the CVSS 9.6 score places it among the most severe bugs Microsoft has ever patched in an AI product. Second, the attack requires no authentication and no user interaction beyond visiting a website: the model itself becomes the execution vector. 

The Exploitability Index Problem 

The SharePoint zero-day (CVE-2026-56164) has exposed a growing credibility gap in Microsoft’s exploitability index. The flaw was rated “Exploitation Less Likely,” yet it was being exploited in the wild before Microsoft shipped the fix. 

Tenable’s Satnam Narang made the connection to broader industry findings: “Anthropic’s Red Team’s own findings for known vulnerabilities (n-days) revealed how fragile this system has become, with its Mythos Preview model being able to produce proof-of-concept exploits for 13 of 14 vulnerabilities that were rated ‘Exploitation Less Likely‘ or ‘Exploitation Unlikely.‘” 

The implication is structural: Microsoft’s exploitability index was designed for human-speed exploitation analysis. AI tools can now produce working exploits for vulnerabilities labelled “unlikely to be exploited” with high reliability. The index no longer predicts real-world risk. 

The Industry Signal 

The 570-figure does not exist in isolation. Multiple major vendors are accelerating their patch cadences: 

• Adobe announced it is moving to twice-monthly security bulletins (2nd and 4th Tuesday), citing AI-driven discovery acceleration. 
• Google shipped over 900 security fixes across its product lines in June 2026 alone. 
• Cisco, Mozilla, and Oracle are all shipping updates more frequently. 
• Apple released over 500 fixes in its July 26.6 update, as noted by security researchers. 

The industry-wide pattern is clear: AI-driven vulnerability discovery is producing a step-change in the volume of known vulnerabilities. The bottleneck is shifting from discovery to remediation. 

What This Means for Defenders 

Five practical takeaways from this Patch Tuesday: 

• Patch volume is the new normal. Expect 200-600 fixes per month from Microsoft going forward. This is not a spike. Allocate patching resources accordingly. 
• Prioritize by active exploitation, not exploitability index. The SharePoint zero-day proves the index lags reality. CISA’s KEV catalogue, vendor threat intelligence, and telemetry from deployed EDR tools are better signals. 
•  Zero-day windows are shrinking but still dangerous. The two actively exploited zero-days were patched within the standard Patch Tuesday cycle. Organizations using extended deployment windows (30+ days) are exposed. 
•  AI products are now attack surfaces. CVE-2026-48561 (Copilot RCE at CVSS 9.6) demonstrates that AI-integrated features introduce new vulnerability classes. Treat AI features as high-risk attack surfaces until proven otherwise.
• Patch Tuesday is no longer a monthly maintenance event. It is a signal about the structural transformation of vulnerability management  driven by AI on both sides of the defensive line. 

AOG Analyst Comment

The 570-patch release is not a Microsoft-specific problem. It is the first visible consequence of AI-driven vulnerability discovery reaching production scale across the industry. Organizations still operating on monthly patch cycles with manual triage will fall behind. The most important defensive investment is not a tool; it is a process redesign that assumes 500+ monthly patches as the baseline, not the anomaly.