The Script Kiddie Shortcut: How AI Is Lowering the Barrier to Cybercrime 

Key Takeaways 

• AI is lowering the skill and resource threshold to launch an attack, expanding the pool of capable threat actors. 

• AI is accelerating attacker workflows rather than replacing human operators. 

• AI complements the existing cybercrime-as-a-service ecosystem by reducing time and technical effort. 

• A July 2026 case confirmed the first purported fully autonomous, agent run ransomware intrusion, though most documented AI use by threat actors still centers on efficiency rather than full autonomy. 

• The greatest near-term impact is likely to be faster, more adaptive campaigns rather than fundamentally new attack techniques. 

Executive Summary 

Artificial intelligence has become one of the most discussed topics in cybersecurity, but much of that discussion centers on whether AI can write malware or generate convincing phishing emails. Those capabilities are real, but they overlook a more consequential shift: AI is changing the economics of cybercrime. 

Over the past decade, cybercrime has matured into a service economy in which threat actors can purchase or lease ransomware, malware, initial access, phishing infrastructure, and stolen credentials. AI extends that model by reducing the time and expertise required for many routine tasks surrounding an intrusion, making threat actors more efficient rather than replacing them. 

For managed service providers (MSPs), the implications are significant. The techniques attackers use to compromise organizations are unlikely to change dramatically, but the speed at which they plan, adapt, and execute campaigns likely will. As AI becomes another tool within the cybercrime ecosystem, defenders should expect familiar threats to evolve faster and place increasing pressure on detection and response. That pressure compounds for MSPs, whose privileged access typically spans dozens of client environments, so a single compromised credential can cascade well beyond one organization. 

AI Didn’t Change Cybercrime Overnight 

Cybercrime has spent years moving toward greater specialization and efficiency, with capabilities that once required deep technical expertise gradually becoming commercialized. Exploit kits lowered the barrier to exploitation. Malware-as-a-service (MaaS) lets operators deploy malware without writing it themselves. Initial Access Brokers (IABs) created a marketplace for compromised networks. Ransomware-as-a-service (RaaS) lets affiliates launch campaigns using infrastructure maintained by someone else. 

Each of these developments reduced the expertise required to participate in cybercrime. Instead of mastering every stage of an attack, operators could acquire individual capabilities as services. AI represents the next step in the evolution, though its effect is different. Rather than making ransomware or malware easier to purchase, AI makes many of the tasks surrounding an intrusion faster to complete. Writing and troubleshooting scripts, modifying publicly available tooling, researching targets, translating content, summarizing technical documents, and drafting phishing lures all require significant time. AI reduces the effort associated with each of these activities by allowing operators to move more efficiently from planning to execution. Open-weight models that can be run locally add a further wrinkle. Commercial AI tools typically screen requests tied to malware, exploits, or attack planning, but that screening runs on the vendor’s own infrastructure. Once an operator downloads the model and runs it on infrastructure they control, there is no vendor in the loop to enforce those protections, so the same restrictions that apply to a hosted commercial tool do not carry over. 

This distinction matters because AI is not replacing established cybercrime business models, but rather only makes them more productive. The result is an ecosystem in which experienced operators can accomplish more with the same resources, while less experienced actors can complete tasks that previously required greater technical proficiency. 

AI Is Becoming Another Tool in the Toolbox 

For several years, discussion of AI in cybercrime remained largely hypothetical, with analysts debating whether AI would eventually write sophisticated malware, discover vulnerabilities, or conduct autonomous attacks. More recent reporting indicates threat actors are incorporating AI into existing workflows in practical ways. 

One of the clearest examples came from the leak of internal communications associated with The Gentlemen ransomware operation. Unlike most threat intelligence, which relies on malware samples, infrastructure, or incident response data, the leaked conversations offered a valuable view into how the operators work. Across multiple reporting organizations, one theme remained consistent. AI had become another productivity tool within the group’s daily operations, supporting code development, troubleshooting, technical research, documentation, and content generation rather than directing attacks independently. None of these tasks alone compromise a network, but together they account for a significant portion of the work required to prepare and sustain an intrusion. 

The Blackpoint Adversary Pursuit Group (APG) has documented multiple campaigns bearing clear indicators of AI assisted development. In one incident, a fake CAPTCHA lure led to a PowerShell based loader carrying telltale signs of AI generation. Those signs included placeholder comments, a fabricated function used purely as filler to waste analysts’ time, and embedded prompt text left behind in Russian. The loader ultimately delivered the DCRat remote access trojan. 

A separate investigation uncovered a previously undocumented framework, tracked as Avalon. Its modular toolset combines credential theft, lateral movement, recovery sabotage, and a ransomware component internally branded CrownX. Assembling it would once have required a seasoned development team, but here it came together with unusual breadth and speed. Blackpoint assessed the framework as showing strong indicators of AI-assisted construction, noting that the rarity of a capability is no longer a reliable signal of the sophistication or resources behind it. 

A third case illustrates the other side of this trend. Attackers are targeting AI tools rather than building with them. A Node.js loader named TaskWeaver was deployed after attackers exploited a SimpleHelp authentication bypass vulnerability. It delivered an information stealer, Djinn Stealer, built with collection rules that specifically target credentials for AI development assistants such as Claude, Gemini, and Codex. Developers often grant these tools standing access to repositories, databases, and cloud accounts. That means stolen tokens can hand an attacker the same downstream reach the developer extended to their AI assistant, extending the breach well beyond the AI tool itself. 

Taken together, these findings point in the same direction as the FakeGit campaign, which similarly showed attackers investing in increasingly convincing software repositories and supporting content designed to establish trust before malware execution. AI is helping attackers perform familiar work more efficiently, and it is also becoming a target, rather than enabling entirely new forms of cybercrime. 

Cybercriminal operations still largely rely on human decision making. Operators choose targets, decide when to pivot, respond to defensive controls, negotiate ransom payments, and adapt campaigns as circumstances change. In most documented cases, AI is not replacing those decisions. It is reducing the manual effort needed to support them, just as it does for legitimate organizations. Threat actors appear to be reaching similar conclusions, applying AI wherever it saves time without surrendering operational control. 

The Real Risk Isn’t Smarter Attacks 

Much of the public discussion around AI focuses on whether it will produce more sophisticated cyberattacks. The more immediate question is whether it lets threat actors work faster, and current reporting increasingly answers yes. 

Cyber intrusions involve far more than exploiting a vulnerability or delivering malware. Campaigns require infrastructure, reconnaissance, malware customization, phishing content, testing, troubleshooting, and continuous adaptation. Individually, many of these tasks are routine. Collectively, they consume considerable time, and this is where AI has had its most tangible effect, compressing many of the delays that previously slowed attacker operations. 

An operator can troubleshoot a PowerShell script in seconds instead of searching documentation for an obscure error. Attackers can rewrite phishing content for different industries or geographic regions within minutes. They can modify public malware more quickly. Research that previously required reviewing dozens of technical resources can often begin with an AI generated summary before analysts validate it against original sources. 

None of these examples fundamentally change how attacks occur. Threat actors continue relying on credential theft, vulnerability exploitation, social engineering, and other well-established techniques. What changes is the amount of time required to prepare, refine, and adapt those techniques throughout an operation, and this compression of attacker timelines carries meaningful implications for defenders. 

Security teams depend on time. It allows analysts to investigate alerts, threat intelligence teams to identify emerging trends, detection engineers to develop new analytics, and incident responders to contain malicious activity before it spreads. As attacker workflows become more efficient, the time available to complete those defensive activities begins to shrink. 

For MSPs, the challenge extends beyond any individual incident. Even modest improvements in attacker productivity become significant when multiplied across hundreds or thousands of customer environments. Organizations should not necessarily expect entirely new attack techniques to emerge overnight, but they should expect existing techniques to evolve and adapt more quickly than they have in the past. 

Looking Ahead: From Assistants to Agents 

Today’s reporting largely shows AI assisting with individual tasks. Operators ask AI to write code, explain errors, summarize research, or generate content before deciding how to use the output. Recent demonstrations of autonomous AI agents suggest the next stage may look considerably different, with models coordinating multiple related tasks with less human direction. Rather than requesting individual scripts or phishing emails, operators may eventually assign broader objectives and let AI determine the intermediate steps. 

Evidence of that transition is already beginning to emerge. In July 2026, researchers demonstrated that an attacker could use the DeepSeek large language model to autonomously identify vulnerable internet facing systems and exploit known vulnerabilities. The model then established persistence and continued operating against compromised servers with minimal human interaction. While the activity relied on publicly known vulnerabilities rather than novel exploitation techniques, it demonstrated how readily available AI models can already coordinate multiple offensive tasks that previously required continuous operator involvement. That distinction is significant because it illustrates that the evolution toward agentic operations is no longer theoretical. 

Additionally in July 2026, researchers at Sysdig documented the first purported fully autonomous, agent run ransomware intrusion, tracked as JADEPUFFER. An AI agent executed the entire attack chain against a live production system with no human operator between the initial exploit and the final data destruction. It chained two known vulnerabilities, harvested credentials, escalated privileges, established persistence, encrypted a database, and left a ransom note correcting its own failed steps along the way. 

None of the underlying weaknesses were new. The intrusion succeeded through an unpatched application and unrotated default credentials, the same gaps that enable human operated intrusions today. The operation fell apart at the last step. The agent never saved its own encryption key and used a placeholder cryptocurrency address, so it could not have produced a payout. 

Government officials have voiced a related concern from a different angle. In July 2026, an FBI deputy assistant director said a preview of Anthropic’s Mythos model had identified previously unknown vulnerabilities in open-source code underlying widely used operating systems, security tools, web infrastructure, and encryption libraries. U.S. export controls briefly applied to the model before officials lifted those restrictions, but the underlying concern persisted. Officials noted that even less capable models could exploit vulnerabilities, similarly, expanding the pool of potential adversaries and complicating attribution. The bureau said it had not yet seen this operationalized at scale but expects that to change. 

It would be premature to conclude that autonomous, agent run intrusions are now mainstream. JADEPUFFER was a single case; its monetization failed outright, and every vulnerability it exploited was already known. It would be equally short sighted to treat it as a one off. The case shows the technical capability for a fully autonomous intrusion already exists, and the most significant shift for defenders may be the disappearance of human decisions between compromise and impact, not new tradecraft. Because an agent that improvises its own infrastructure leaves few durable indicators, detection will likely depend more on behavior than signatures going forward. Adoption is more likely to happen gradually than through a single breakthrough, but that breakthrough has already occurred, and the shift is underway. 

Conclusion 

Artificial intelligence is unlikely to change the fundamental objectives of cybercrime. Threat actors will continue pursuing financial gain, espionage, and disruption using many of the same techniques defenders have tracked for years. What changes is how efficiently threat actors can pursue those objectives. 

AI is becoming another capability within an already mature cybercrime ecosystem, cutting the time and expertise those routine tasks once required. Experienced operators can execute campaigns more efficiently, while less experienced actors gain assistance with technical work that previously represented a barrier to entry. The population capable of running an intrusion end to end is larger than it was even two years ago, which raises the question our upcoming webinar takes up directly. Is the era of the script kiddie back, just with a far better toolkit? 

For MSPs, this means preparing for an environment in which familiar threats evolve more rapidly, rather than an entirely new generation of AI driven attacks. The organizations best positioned to defend against this shift will be those able to detect malicious activity quickly, investigate suspicious behavior, and adapt alongside an adversary that is steadily becoming more efficient. 

The question is no longer whether AI will become part of cybercrime. That transition is already underway. The more important question is how defenders will adapt as attackers continue finding ways to move faster with the help of their newest assistant. 

Join Our Live Webinar ‘Is the Era of the Script Kiddie Back?’  

We will unpack this shift in more detail, and what it means specifically for MSPs facing a larger and more capable pool of attackers, in our live webinar ‘Is the Era of the Script Kiddie Back?‘ Register here.

Wednesday, August 26, 2026 

2:00 to 3:00 PM ET/11:00 AM to 12:00 PM PT 

Featuring Wes Spencer, Co-Founder at Empath, Erin Whitmore, Head of Adversary Pursuit Group at Blackpoint Cyber, and Wilfredo Santiago, Chief Security and Trust Officer at Blackpoint Cyber.