Beyond the Hype: Why AI Is Making Cyberattacks Less Effective, Not More Dangerous

For the better part of three years, the cybersecurity industry has been gripped by a singular narrative: the rise of Artificial Intelligence is destined to arm cybercriminals with god-like capabilities. We have been repeatedly cautioned that threat actors are leveraging Large Language Models (LLMs) to bridge the gap between amateur status and professional-grade sophistication, enabling them to execute faster, more lethal, and more resilient campaigns. However, according to recent analysis from the front lines of digital defense, this widely accepted industry wisdom may be more rooted in marketing fiction than operational reality.

At the IRISSCON 2026 conference held in Dublin on November 11, Selena Larson, a principal threat researcher at Proofpoint, delivered a candid assessment that challenged the prevailing industry dogma. While the common refrain suggests that AI allows attackers to operate with the same resources as elite security professionals, Larson argued that the empirical data tells a much more nuanced—and perhaps surprising—story. Far from elevating the threat landscape, the over-reliance on generative AI tools has, in many instances, resulted in a measurable decline in the quality and efficacy of malicious campaigns.

The Rise of "AI Slop" in the Attack Chain

Larson’s core thesis is that the industry has conflated the speed of output with the quality of execution. While AI platforms undoubtedly allow threat actors to scale their operations and automate the generation of content, this push for efficiency has introduced a phenomenon Larson terms "AI slop." This describes the proliferation of error-ridden, nonsensical, or disjointed artifacts that appear throughout the modern attack chain.

Proofpoint’s research has uncovered a consistent pattern of dysfunction within AI-generated malware. Larson noted that her team frequently encounters malicious code that is fundamentally broken. In some cases, the comments within the code explicitly state an intention to perform one action, while the functional logic of the script executes something entirely different. This misalignment suggests that threat actors are using AI as a crutch rather than a tool, generating code they may not fully understand or have the expertise to audit before deployment.

The impact of this trend is perhaps most visible in the realm of phishing and business email compromise (BEC). Larson highlighted the activities of a specific threat actor tracked by Proofpoint as TA4903. Historically, this group was known for its disciplined approach, successfully impersonating high-profile US government entities to target various sectors with carefully crafted, sophisticated lures. However, since the group began integrating AI tools into their workflow, the quality of their campaigns has deteriorated significantly.

The "AI-generated" nature of these new lures is often glaringly obvious. Analysts have observed messages that suffer from severe structural failures—subjects that do not match the content, attachments that bear no relation to the body text, and, in some instances, messages that are sent entirely blank. These blunders are not merely cosmetic; they serve as red flags that make the attacks significantly easier for both automated security filters and human users to detect and flag.

Furthermore, the shift toward using AI-driven coding tools for the creation of phishing websites has produced what some have dubbed "vibe coded" pages. These sites are often simplistic, characterized by repetitive design elements, generic color palettes, and standardized formatting. Because the AI is relying on training data that mirrors common web patterns, the resulting phishing sites lack the unique, convincing touches that human attackers once employed to bypass suspicion. These generic, cookie-cutter templates are inherently easier for security teams to identify as malicious.

The Paradox of Over-Polished Phishing

One of the more counterintuitive findings presented by Larson concerns the use of AI to "perfect" the language used in phishing emails. For years, security awareness training has taught employees to look for poor grammar, spelling mistakes, and awkward phrasing as hallmarks of a fraudulent message. Consequently, attackers have turned to AI to ensure their emails are grammatically flawless and linguistically precise for the target region.

However, Larson suggests that this obsession with "professionalism" is actually a tactical disadvantage. Real human communication—even in a business context—is rarely perfect. People make typos, use informal syntax, and often fail to write with a polished, corporate tone. By using AI to create perfectly synthesized, error-free messages, attackers are inadvertently stripping their lures of the "human element" that makes them appear authentic. This hyper-polished output often stands out to observant recipients as unnatural, ironically making the emails less believable than the sloppy, human-authored attempts of the past.

Vulnerability Exploitation and the "BlueMoon" Example

The limitations of AI are not confined to social engineering; they are equally apparent in the technical exploitation of software vulnerabilities. While it is true that AI can assist in identifying potential flaws in software, the transition from discovery to a functional, weaponized exploit chain is a different challenge entirely.

Larson pointed to the "BlueMoon" exploit chain, which was first observed in August 2026. This campaign involved the chaining of three distinct vulnerabilities within the Microsoft Chrome browser. The threat actors involved utilized LLMs to help draft the exploit code, pulling information from official patch release notes. Despite the speed at which the actors were able to assemble the chain, the final product was highly flawed. The AI-generated exploit failed to perform reliably, often triggering erratic behavior and creating a high volume of system noise. From a defensive perspective, this "noise" was a gift; it rendered the exploit significantly easier for security operations centers to monitor, isolate, and neutralize.

Redefining the Defensive Mandate

The overarching message from Larson’s presentation was one of pragmatic reassurance. While the media and security vendors have spent the last few years breathlessly detailing the "AI-driven apocalypse," the reality is that the fundamental nature of cyber defense has not undergone a radical transformation.

Larson emphasized that AI should be viewed strictly as a productivity tool—one that is entirely dependent on the competence of its operator. If the operator lacks the fundamental knowledge of how to construct a sophisticated attack, no amount of AI-generated code or text will compensate for those gaps in expertise. If anything, the technology provides a false sense of security to the attacker, leading to careless mistakes and a reliance on automated outputs that are inherently flawed.

For security professionals, this means that the core defensive priorities remain unchanged. The threat landscape continues to evolve, but the "AI revolution" has not fundamentally altered the underlying principles of risk management, vulnerability patching, and employee awareness. Defenders must continue to focus on the basics: maintaining visibility, monitoring for anomalous activity, and ensuring that security controls are robust enough to catch both sophisticated human-led threats and the "sloppy" automated attacks produced by less skilled adversaries.

Ultimately, Larson’s analysis serves as a vital reminder that technology is a force multiplier, not a replacement for human intellect. AI may make the process of creating an attack faster, but it does not make the attacker smarter, more patient, or more precise. As the novelty of AI-generated threats continues to wane, the cybersecurity industry may find that the most dangerous threats remain those driven by the deliberate, strategic, and human-led efforts that have defined the landscape for decades. Attacks are, and will remain, attacks; the tools used to launch them may change, but the defense remains constant.