A new report reveals a sharp rise in malicious emails, with attackers increasingly using AI-driven phishing and trusted infrastructure to bypass defenses.
The Q3 2025 Email Threat Trends Report recorded a 13% year-over-year surge in detected malicious emails. Researchers analyzed 1.8 billion emails during the quarter, identifying more than 234 million spam messages, of which 26 million contained active threats. The findings point to increasingly sophisticated credential-harvesting and social-engineering campaigns targeting email platforms such as Outlook and Gmail.
Sandboxing technology intercepted nearly 150,000 new malicious attachments and over 67,000 previously unseen malicious links that slipped past traditional filters, evidence that attackers are refining their methods to outpace static defenses.
Credential theft was the primary objective of most phishing campaigns, with over 90% focused on stealing Outlook and Google account credentials. Attackers exploited these widely used platforms for both personal and enterprise targets, increasing their reach and success rates.
Malicious links served as the primary attack vector in 65% of cases, followed by PDF-based attachments, which were used in 31% of campaigns. Meanwhile, 63% of business email compromise (BEC) attacks involved impersonating senior executives, often diverting conversations to encrypted or unmonitored messaging applications, such as WhatsApp.
The report also noted that AI-generated emails now make up 57% of BEC attempts, an indication that generative tools are being used to create convincing, grammatically correct phishing lures at scale.
Researchers stated that threat actors use trusted services and infrastructure to avoid detection. About one-third of malicious emails originated from compromised legitimate accounts, while another third used free relay providers such as SendGrid and Amazon SES to distribute harmful content.
Attackers also rely on disposable domains, typo-squatting, and open redirects to mask their true origins. English remains the dominant language in 87% of samples, suggesting campaigns are primarily directed at global business hubs.
New tactics, including the misuse of Apple’s TestFlight to distribute iOS malware and adaptive delivery of PDQ RAT trojans, show how attackers are continuously changing their methods.
Attackers are no longer relying on shady domains or poorly written scams; they’re exploiting trusted infrastructure like Outlook, Gmail, and cloud relay services to blend in with legitimate traffic. Even advanced filters struggle to catch threats that originate from known platforms or hijacked accounts, especially when the phishing content itself is generated by AI. The result is a new level of email deception that looks and feels authentic to both users and standard security systems.
Since email remains the most exploited attack vector, relying on traditional filtering is no longer enough. Paubox Inbound Email Security provides detection that goes beyond static signature checks, identifying abnormal communication patterns even when messages come from legitimate services. For healthcare and enterprise environments where Outlook and Gmail are core tools, Paubox helps stop credential theft, BEC scams, and AI-generated phishing before they ever reach the inbox.
Their widespread use across both personal and corporate environments gives attackers access to a vast number of potential victims and interconnected networks once credentials are compromised.
AI allows attackers to rapidly generate personalized and convincing messages, reducing spelling or tone errors that traditionally helped users spot phishing attempts.
Attackers use legitimate mail services to distribute malicious messages because emails from trusted infrastructure are less likely to be flagged by filters.
Open redirects occur when legitimate websites are manipulated to redirect users to malicious pages, helping phishing links appear safe and bypass standard URL checks.
By combining behavioral analytics, AI-based anomaly detection, and sandboxing, organizations can identify and neutralize threats that static filters miss.