Where AI Meets Adversary: The Future of Continuous Pentesting

Written by Travis DeForge, Director of Cybersecurity at Abacus Group

Most businesses we work with have begun their journey of embracing and integrating artificial intelligence (AI) into their workflows, developing new capabilities, and augmenting staff to increase efficiency. Simultaneously, vendors are leveraging new capabilities, particularly those related to large language models (LLMs) and agentic AI, to enhance their products at a breakneck pace. Cyber criminals have emerged as both early adopters and, unfortunately, innovators in the use of AI to target and extort businesses at an incredible pace. 

I’ve written previously about the effectiveness of generative AI in facilitating social engineering by malicious actors, who produce high-quality phishing emails and even leverage deepfakes to trick end-users. Still, in this installment, we will dig into an even more insidious new threat – Agentic Ransomware. 

According to this year’s Global Threat Report from CrowdStrike, the average time to move laterally within a network, which measures how long it takes a cybercriminal to compromise additional devices following initial access, has decreased on average to 48 minutes. But, most alarmingly, the fastest recorded breakout was a mere 51 seconds. This speed suggested a significant augmentation of traditional capabilities, but this fear was later reinforced by the discovery of the first known AI-powered ransomware, named PromptLock, in August 2025. First reported by ESET, PromptLock utilizes an open-source AI model to generate custom code that can enumerate file systems, inspect files, exfiltrate data, encrypt files, and even destroy data, all with minimal user input. 

Why Agentic Ransomware Changes the Game

Agentic ransomware, such as PromptLock, represents a shift from static, pre-coded malware to adaptive, on-demand attack engines. These tools can write and execute custom code in real time, making them harder to detect and faster to act.

As these offensive capabilities evolve, defending against them will require equally advanced, adaptive approaches; ones that go beyond annual checkups to continuous, intelligence-driven testing that mirrors the sophistication of the threat.

This philosophy was the impetus for Abacus Group to develop a way to leverage agentic AI technology to enhance proactive security testing in a way that can safely be as thorough as traditional testing, while increasing the frequency of assessments. 

Introducing AI-Enabled Continuous Penetration Testing

Abacus Group now offers a cutting-edge adversarial simulation service designed to emulate the tactics, techniques, and procedures of a sophisticated threat actor targeting your network. Our methodology combines the precision and scalability of agentic AI technology with the strategic insight of senior Red Team penetration testers. This hybrid model ensures that every test is both technically rigorous and contextually intelligent. To ensure parity with conventional penetration testing standards, our Red Team overlays expert analysis atop AI-driven reconnaissance and exploitation, delivering depth, nuance, and actionable intelligence.

By conducting these simulations on a routine basis (as opposed to annually), organizations gain continuous visibility into emerging risks, far surpassing the static insights of traditional annual testing.

In a world where attackers move in seconds, this cadence transforms security from a yearly checkbox into a living, breathing defense posture that evolves as fast as the threats themselves. It’s the difference between catching a spark before it becomes a wildfire, identifying and addressing vulnerabilities while they’re still in their infancy.

Ready to strengthen your defenses with continuous, intelligence-driven security testing? Learn more about our AI-Enabled Penetration Testing service and contact us today to safeguard your firm against evolving threats.