Attackers weaponize trusted AI coding assistants and remote tools

The shift toward trusted software

Cybersecurity defenses have historically relied on a foundation of trust. Organizations typically trust AI coding assistants because they are on approved software lists, and they trust remote monitoring and management (RMM) tools because IT departments install them with trusted signatures.

However, a new report from Elastic Security Threat Command reveals that adversaries are now flipping this foundation on its head. Instead of trying to defeat security mechanisms with new malware, they are weaponizing the inherent trust placed in the enterprise stack to gain access.

By using legitimate products, attackers can execute commands much faster than they could with traditional methods. For example, out of 23 confirmed RMM abuse findings this quarter, ScreenConnect was the product named most often by threat actors.

Speed of AI-driven intrusions

The use of AI agents has significantly accelerated the pace of cyberattacks. In one specific instance, an attack driven by an AI agent resulted in roughly 17,600 reconstructed actions that were compressed into a window of just four days.

This speed is further amplified when attackers use legitimate workflows. In a confirmed healthcare finding, the transition from a paste-and-run lure to delivery via a Microsoft Software Installer (MSI) took roughly 16 seconds because the software was already trusted.

Attackers are also exploiting the consent loophole to bypass multifactor authentication (MFA). By using phishing kits that exploit legacy authentication, they can move from an initial infection to logging into a cloud tenant in an average of only eight minutes.

Why signatures are no longer enough

Traditional signature-based detection is becoming less effective against these threats. Because the software being used is signed and legitimate, a security system cannot simply block the application based on its identity or file hash.

To counter this, defenders must move toward system-level behavioral outcomes. This means monitoring for specific unauthorized actions, such as memory tampering or proxy tunneling, regardless of which specific application is initiating the activity.

This shift creates a significant challenge for security teams because it generates a much higher volume of alerts. Distinguishing between a developer using an AI assistant for a valid task and an attacker using it for a malicious one requires much deeper visibility into the environment.

New strategies for detection

Detecting these threats requires user and entity behavior analytics (UEBA) to establish a baseline of what normal activity looks like. Security teams need to correlate data across different domains, including containers, cloud control planes, and legitimate user accounts.

Because the window between abusing trust and gaining access is shrinking, manual response is often too slow. The report suggests that practitioners must prioritize automation to keep up with machine-speed attacks that occur in seconds or minutes.

Elastic Workflows is one tool mentioned that attempts to solve this by combining telemetry context with deterministic rules and AI reasoning. This approach aims to filter out benign activity automatically so that analysts only see high-fidelity threats that require immediate human intervention.

The evolving threat landscape

The rapid weaponization of trusted software means that annual security reports are no longer sufficient for keeping up with changes. Elastic has moved to a quarterly cadence to better track the evolution of AI-assisted offense and other emerging tactics.

The goal of these reports is to provide defenders with the exact mechanics of an intrusion rather than just raw numbers. Understanding the specific steps an attacker takes—such as how they move from an infostealer to a cloud tenant—is more useful for defense than simply knowing how many attacks occurred.

As attackers continue to exploit the gray areas of identity and software trust, the focus of security will continue to shift. Organizations must move away from binary 'safe' or 'unsafe' labels and instead hunt for the abuse of the tools they already use every day.