Microsoft Threat Intelligence says it detected approximately 7.3 billion email-based phishing threats between July and September 2026. Nearly three billion were detected in September alone, up 41 percent from August, according to the company's quarterly report published on 5 November. These figures represent Microsoft's detections, not a worldwide total.
The increase was not uniform across attack techniques. QR-code phishing remained well below its early-year peak, while CAPTCHA-gated phishing began to recover. Microsoft detected 3.1 million such attacks in September, almost twice July's low. SVG attachments were central to the initial rebound and accounted for 49 percent of CAPTCHA-gated attacks in September.
Microsoft also identified approximately 129 million malicious messages spoofing the sender domains of business services covered by its analysis. According to the report, many were delivered despite failing authentication and being identified as malicious because organization- or user-level exceptions overrode security protections. Microsoft explicitly cautions that the figures do not establish that recipients opened the messages or that accounts were compromised.
One campaign highlighted in the report, observed on 19 August, targeted more than 10,000 users across over 4,800 organizations, almost exclusively in the United States. Personalized salary-review documents and QR codes led to a device-code phishing flow. Completing it could give an attacker a valid session token without exposing the password or triggering multifactor authentication again. A separate campaign on 21 September used purchase-order and payroll lures to distribute Agent Tesla malware.
Among its recommendations, Microsoft advises verifying senders and requests, reviewing overly broad security exceptions, and using passwordless or phishing-resistant authentication such as FIDO security keys. Users of encrypted mailboxes should likewise remain cautious around sign-in links, QR codes and unexpected attachments: encryption alone does not prevent credentials or sessions from being surrendered through a fraudulent page.