IR trends Q3 2026: Actors impersonate IT help desks in vishing attacks as ransomware activity increases

Phishing remains the leading initial access vector, though its prevalence decreased from over 50 percent in the prior quarter to just over one-third (37 percent) of Cisco Talos Incident Response engagements. The nature of these attacks has evolved, incorporating voice phishing (vishing) combined with IT desk impersonation to trick victims into granting remote access via legitimate Remote Monitoring and Management (RMM) tools. Exploitation of public-facing applications ranks as the second most common vector at one-third (33 percent). Ransomware and pre-ransomware incidents rose to 27 percent in Q3 2026, impacting engagements involving sophisticated RaaS groups like The Gentlemen and Rhysida, alongside emerging variants such as VYPR. These threat actors frequently employ evasion techniques, notably disabling Endpoint Detection and Response (EDR) systems by exploiting vulnerable drivers or manipulating Windows Defender Application Control (WDAC).
Cyber espionage targeting the telecommunications sector shows prolonged activity attributed to China state-sponsored actors, including Salt Typhoon. These intrusions often leverage valid credentials and known vulnerabilities in web-based management interfaces. Across all surveyed incidents, organizational security weaknesses are consistently noted, with insufficient logging and lack of Multi-Factor Authentication (MFA) being among the most frequently observed flaws. Furthermore, the use of Artificial Intelligence is emerging as a dual vector: both an attack surface target (e.g., credentials scraped from AI assistant keychains) and a weaponized tool in insider threats.

The persistent narrative arc here is one of escalating sophistication layered upon foundational human and technical negligence. The shift in initial access methods—from pure link-dropping phishing to socially engineered remote takeover via vishing paired with RMM access—suggests a refinement in the *delivery* mechanism, not necessarily the core exploitability. Attackers are moving from being purely external hackers to being operational proxies who exploit established organizational processes (like help desk functions) for initial foothold.
The most critical conceptual failure highlighted is that threat actors routinely treat sophisticated security measures (EDR, MFA) as mere speed bumps rather than insurmountable barriers; disabling defenses through driver exploitation or configuration manipulation demonstrates a fundamental belief in the attacker's ability to achieve 'operational impunity.' This suggests an assumption within both criminal operations and defensive planning: that high-level tools can always be circumvented by targeting low-level trust mechanisms (drivers, administrative interfaces).
The dual nature of AI—as a target for credential scraping (the digital "password" of the AI ecosystem) and as a potential weaponized utility—signals a paradigm shift where organizational risk is no longer limited to traditional file paths or network perimeter breaches. The cost borne by organizations appears heavily concentrated in the lack of systemic hygiene: insufficient centralized logging means that when defenses are breached, accountability evaporates into unrecoverable gaps.
Patterns detected: ARC-0045 Assumption Bias (Underestimating systemic weakness resilience), ARC-0017 Externalization of Blame (Attribution to actor sophistication rather than infrastructure flaw)
The cost is borne by the organization whose security posture can be mapped down to a failure in documentation or MFA implementation. Second-order consequences include the normalization of high-stakes cyber operations within standard business workflows, blurring the line between "administrative activity" and malicious persistence. What underlying cultural assumption permits such porousness—the belief that operational necessity must supersede security discipline? If RMM tools are essential for operation, how can organizations realistically segregate administrative access rights to prevent them from becoming privileged vectors for external intrusion? How does the rise of AI force us to redefine "trust" within an enterprise environment when credentials securing knowledge platforms are targets themselves?