SaaS Supply Chain Attacks and Shadow AI
Introduction
The Monthly Threat Report by Hornetsecurity brings you monthly insights into M365 security trends, email-based threats, and commentary on current events in the cybersecurity space. This edition of the Monthly Threat Report focuses on industry events and content from the month of August 2025.
Executive Summary
Threat Overview and Industry Events
Microsoft Locks Down .onmicrosoft Domains
In late August, Microsoft announced a major policy shift for its legacy .onmicrosoft.com domains. Historically, tenants could use these default cloud-issued domains for outbound email, which attackers increasingly abused for phishing and spam. Beginning this fall, outbound traffic from these domains will be throttled to 100 external recipients per day per organization. This effectively neuters onmicrosoft.com domains for bulk communication. Microsoft is making it clear: if you want to send business email, you need to verify and configure your own domain. Thankfully the impact here should be low as most security teams have looked at locking down this attack vector some time ago. That said, many organizations still use onmicrosoft.com domains for subscription-related communications in M365, which SHOULD be covered by the newly imposed limit.
Why does it matter?
This is less about throttling convenience and more about forcing secure identity practices. Organizations that haven’t fully migrated to custom domains will be impacted, and threat-actors lose yet another free ride for impersonation.
Salesloft / Drift OAuth Supply Chain Breach
A supply chain compromise involving Salesloft’s Drift integration rattled a number of Salesforce customers this month. According to The Hacker News, OAuth tokens associated with Drift were stolen, granting unauthorized access to sensitive Salesforce data. Zscaler, Google, and others confirmed exposure, showing how a single SaaS integration can ripple across the ecosystem. The incident really drives home the fact that security is VERY difficult in the modern cloud ecosystem and software as a service apps. SaaS-to-SaaS trust chains are now just as juicy a target as on-prem software supply chains.
Why does it matter?
OAuth integrations are everywhere, often with elevated privileges and little oversight. A single compromise at the vendor level can cascade into dozens of high-value environments, bypassing traditional security controls, and threat-actors know this.
August 2025 Patch Tuesday – 107 CVEs
Microsoft’s August Patch Tuesday advisory was a big one with 107 vulnerabilities closed, with 13 critical flaws and one already exploited zero-day. Patches spanned Windows, Office, and Azure components, with particularly bad remote code execution opportunities in Excel and SharePoint. The scope makes it one of the year’s heavier updates and reinforces the trend of sprawling patch loads hitting admins mid-quarter.
Why does it matter?
The sheer volume combined with critical RCEs raises patch fatigue risk. Organizations that lag behind could leave massive attack surface areas exposed for opportunistic exploitation.
WinRAR Zero-Day Exploited in the Wild (CVE-2025-8088)
Archival software continues to be a favorite playground for attackers, and sometimes the target as well, it seems. Researchers confirmed that WinRAR was hit with a path traversal bug allowing remote code execution via crafted archives. Dubbed CVE-2025-8088, the flaw has already been leveraged in real-world campaigns. The vendor responded with an update to WinRAR 7.13, but with the software’s massive install base across a diverse array of machines, patch adoption is expected to be slow.
Why does it matter?
WinRAR remains ubiquitous across IT ecosystems, and it’s not often thought of in terms of as part of the traditional attack surface. That combination makes this an evergreen attack vector until organizations enforce patching.
Cisco Secure Firewall FMC – Critical RCE
Cisco disclosed a maximum severity flaw in its Secure Firewall Management Center (FMC) (CVE-2025-20265). The vulnerability exists in the RADIUS subsystem and could allow an unauthenticated attacker to execute arbitrary commands on the underlying system. Cisco shipped fixes and published mitigations, but warned exploitation would be trivial for a motivated actor. The FMC platform manages firewall deployments at scale, raising the stakes for anyone lagging on the patch.
Why does it matter?
Management platforms represent “keys to the kingdom” access. A compromise here doesn’t just expose a device, it risks every downstream firewall and policy in the enterprise, making them a prime target for attackers.
From the Hornetsecurity Blog
SharePoint & Exchange Vulnerabilities – Déjà Vu All Over Again
Few things cause CISOs to sweat quite like an unpatched Microsoft server sitting on the perimeter. This summer, SharePoint Server took center stage after the ToolShell flaw, first revealed at Pwn2Own, quickly snowballed into real-world exploitation. Attackers wasted no time: Chinese-linked groups including Linen Typhoon, Violet Typhoon, and Storm 2603 weaponized the bug to steal IP, conduct espionage, and even deploy ransomware via Group Policy. High-profile victims included the U.S. Department of Homeland Security, the NIH, and the National Nuclear Security Administration, showing that this wasn’t just a nuisance vulnerability but a global security event.
And just as defenders started to catch their breath, Black Hat 2025 brought more bad news: a critical Exchange Server flaw (CVE-2025-53786) resurfaced, leaving thousands of hybrid deployments exposed. Once again, exploitation chains highlighted how on-premises holdouts remain a soft spot in enterprise security. Federal agencies were even slapped with an emergency directive to patch immediately, evidence that regulators see these incidents as systemic risks, not just IT headaches.
DeepSeek AI – The Hype, The Holes, and the Hard Truths
DeepSeek’s meteoric rise to the top of app store charts has been matched only by the speed of the backlash. Security researchers uncovered unprotected databases, exposed plaintext chat logs, API keys, and backend details, making it clear that the platform shipped without even the most basic security hygiene. Couple that with a privacy policy that openly collects keystroke biometrics, device info, and chat histories—all stored in China, and you have a recipe for global regulatory alarms. Italy’s data protection authority has already banned the service, and other EU regulators are circling.
What makes this case especially thorny is the illusion of anonymity. Keystroke biometrics can uniquely identify users, effectively undoing the privacy benefits of VPNs or pseudonymous accounts. Pair that with DeepSeek’s failure of over half of Qualys’ jailbreak tests, and the tool becomes a double-edged sword: powerful AI capabilities on one side, a glaring security liability on the other. For enterprises, the lesson is clear: enthusiasm for AI adoption MUST be balanced with rigorous vendor risk assessments, especially when data sovereignty and compliance hang in the balance.
Predictions for the Coming Months
Monthly Recommendations