The landscape of cyber-adversarial activity is undergoing a significant transformation, moving away from simple credential harvesting toward the sophisticated manipulation of trusted administrative tools and artificial intelligence systems. According to recent intelligence reports published by security firms Fortra and Barracuda, threat actors are leveraging legitimate Remote Monitoring and Management (RMM) software and injecting deceptive instructions into AI-powered e-mail assistants to bypass traditional security perimeters. These findings underscore a critical shift in the threat model: the weaponization of the very productivity tools designed to streamline business operations.
The Rise of RMM-Enabled Phishing Campaigns
Fortra’s analysis reveals a staggering 475% increase in phishing attacks incorporating RMM tools during the first three quarters of 2026 compared to the entirety of 2025. This surge is not merely a statistical anomaly but a strategic pivot by organized cyber-criminal groups, particularly those targeting North American financial institutions.
Remote monitoring and management software, such as AnyDesk or TeamViewer, are indispensable assets for IT departments, providing the capability to troubleshoot systems, deploy software updates, and manage networks from a distance. However, the inherent power of these tools—unrestricted access to a user’s desktop—is exactly what makes them attractive to malicious actors.

The typical attack vector follows a highly structured social engineering methodology. It often begins with an unsolicited communication—via phone, e-mail, or SMS—warning the victim of an urgent security issue with their bank account. The victim is directed to a fraudulent website that mimics the interface of a legitimate financial institution, complete with a live customer support chat interface.
Once the victim engages with the fake chat, the malicious script performs a system reconnaissance check. If the visitor’s operating system is compatible, the site triggers an automatic download of a remote management application. The attacker then assumes the role of a helpful support technician, guiding the victim through the installation process. By coaxing the user into providing the connection key, the attacker gains persistence on the device. This "support" interaction essentially grants the adversary a persistent backdoor, allowing them to monitor keystrokes, exfiltrate sensitive financial data, and install secondary payloads such as ransomware or persistent malware.
A Chronology of Escalation
The emergence of these RMM-based attacks can be traced back to the third quarter of 2025, when researchers first identified a localized spike in activity targeting U.S. commercial banking users. By the first quarter of 2026, the activity had reached a critical peak. During this period, attackers standardized their infrastructure, relying heavily on Firebase-hosted links to deliver RMM executables.
In response to the identified abuse, AnyDesk implemented restrictions on direct downloads originating from Firebase domains. While this proactive measure succeeded in curbing the initial wave of attacks, it served only as a temporary deterrent. Threat actors quickly demonstrated their adaptability, shifting their delivery mechanisms to alternate hosting services and embedding download redirects within obscured webpage elements. This cycle of attack and mitigation illustrates the "cat-and-mouse" nature of modern cybersecurity, where vendors are forced to balance the accessibility of their tools with the security of the end-user.

Weaponizing AI Assistants: The New Frontier
While RMM-based attacks focus on manual manipulation, the report from Barracuda highlights a more futuristic, albeit equally dangerous, threat: the exploitation of AI-driven e-mail assistants. As organizations integrate AI tools to summarize long e-mail threads and prioritize communications, they have inadvertently created a new attack surface.
Barracuda’s findings indicate that attackers are now embedding "hidden instructions" within e-mails—content that is invisible to the human reader but legible to an AI processor. By utilizing zero-width characters, hidden text, or malicious metadata, attackers can manipulate the summary generated by an AI assistant. For instance, an invoice phishing email might contain hidden commands directing the AI to flag the message as "urgent" or "trusted," or to instruct the assistant to incorporate fraudulent payment instructions into the summary presented to the employee.
"Protecting the inbox is no longer enough," noted Guruprasad Kenja, a threat analyst at Barracuda. "We are seeing a convergence of traditional social engineering and prompt injection attacks that leverage the trust employees place in automated processing tools."
Broader Implications for Enterprise Security
The confluence of these two trends—RMM exploitation and AI manipulation—represents a significant challenge for IT and security operations centers (SOCs). Traditional security controls, which rely on blacklisting known malicious files, are becoming increasingly ineffective. Because RMM tools are legitimate, signed software, security suites often fail to distinguish between an authorized administrator and a malicious actor using the same tool.

Furthermore, the automation of internal processes via AI means that security teams must now perform "post-inbox" analysis. It is no longer sufficient to secure the perimeter; organizations must now scrutinize how automated agents process data. The implications of these threats are profound:
- Identity and Access Management (IAM): The use of RMM tools suggests that even with multi-factor authentication, an attacker can gain control over an authenticated session, effectively bypassing the security gate.
- AI Governance: Organizations must implement strict controls on AI assistant permissions. The "human-in-the-loop" requirement for sensitive actions, such as changing vendor payment details, has become a non-negotiable security requirement.
- Endpoint Visibility: The ability of attackers to hide in plain sight necessitates more robust endpoint detection and response (EDR) solutions that monitor behavioral patterns rather than just file hashes.
Expert Recommendations and Future Outlook
Industry experts are calling for a multi-layered defensive strategy to combat these emerging threats. For RMM-based risks, both Fortra and Barracuda emphasize the importance of maintaining a strictly enforced allowlist for administrative tools. If an application is not authorized for standard business operations, it should be blocked by default at the endpoint level. Additionally, financial institutions are encouraged to launch consumer awareness campaigns to help users identify the hallmarks of fraudulent support interactions.
For AI security, the recommendations are more technical. Organizations are advised to:
- Sanitize inputs: Remove all hidden or non-visible content from e-mails before they are passed to an AI processing engine.
- Prompt Injection Detection: Deploy specialized security filters designed to identify and block common prompt injection patterns.
- Human-in-the-Loop: Ensure that any summary or recommendation generated by an AI assistant regarding financial or sensitive operational data is verified by a human operator before action is taken.
The findings from these reports signal a maturation in cyber-criminal tactics. As businesses continue to adopt remote management software and AI assistants to increase productivity, they must simultaneously evolve their security postures. The era of simple, signature-based detection is effectively over. The future of enterprise defense lies in behavioral analysis, rigorous verification of automated outputs, and the realization that trust, once extended to a tool or an AI, can be exploited by an adversary.

As the industry looks toward the remainder of 2026 and beyond, the focus must shift from reactive patching to proactive architectural changes. Whether it is a fake support technician controlling a computer via AnyDesk or a hidden prompt manipulating an AI summary, the common thread is the exploitation of trust. By narrowing the scope of what is considered "trusted" and demanding human oversight for high-impact decisions, organizations can begin to mitigate these sophisticated threats. Ultimately, the security of the modern enterprise depends not just on the software it uses, but on the vigilance with which it monitors the interactions between humans, machines, and the automated systems that connect them.








