For the past several years, the cybersecurity industry has been dominated by a singular, persistent narrative: the rise of Artificial Intelligence will fundamentally transform the threat landscape, turning even the most novice cybercriminals into sophisticated, state-level operators. From boardrooms to security operations centers, the prevailing wisdom has been that AI would grant attackers unprecedented capabilities, allowing them to scale their operations, automate complex exploits, and outmaneuver the most robust defensive systems.
However, as the dust settles on early deployments of these technologies, evidence is mounting that the reality of AI-driven cybercrime is significantly less ominous than the marketing hype suggests. According to new research presented at the IRISSCON 2026 conference in Dublin, Ireland, on November 11, the integration of AI tools into the malicious attack chain has not necessarily led to a more lethal threat landscape. In fact, for many threat actors, the reliance on these tools has inadvertently rendered their campaigns less effective, more detectable, and prone to catastrophic failure.
The Myth of the Super-Powered Attacker
Selena Larson, a principal threat researcher at Proofpoint, offered a sobering assessment of the current state of AI-assisted cybercrime during her keynote presentation. Larson, who has spent months analyzing the methodologies of various threat groups as they experiment with Large Language Models (LLMs) and automated coding tools, challenged the industry-wide consensus that AI serves as a "force multiplier" for bad actors.
"We’ve all been told since 2023 that AI will make attackers better, faster, and stronger," Larson stated. "The prevailing narrative is that they will have access to the same tools and resources as the best developers, security professionals, and vulnerability researchers in the world, and will automatically become better at hacking. This is marketing nonsense; the data just doesn’t back up this assertion."
Larson’s research suggests that while AI tools are undeniably capable of increasing the speed and volume of certain attacks, they are failing to provide the nuance and precision required to execute truly dangerous, high-impact campaigns. By leaning too heavily on automation, threat actors are falling into a trap of their own making, introducing a new type of vulnerability into their own processes.
The Rise of AI "Slop" in the Attack Chain
The term "AI slop"—a colloquialism describing low-quality, AI-generated content—has increasingly become a defining characteristic of modern cyberattacks. While the term is often used in the context of generic social media content, Proofpoint’s research indicates that it is now bleeding into the core components of malicious campaigns, including malware development and phishing infrastructure.
Larson highlighted that Proofpoint researchers have encountered a surge in malware samples that are either partially or fully generated by AI, often with disastrous results. These samples frequently contain code that is fundamentally broken, logically incoherent, or completely at odds with its intended function.
"We frequently run into malware where the code doesn’t actually work," Larson explained. "There are comments within the code that express an intent to perform one action, yet the malware itself is executing something entirely different. It is a symptom of threat actors over-relying on a tool they don’t fully understand or audit."
This loss of quality control is perhaps most visible in the realm of phishing. The threat group tracked by Proofpoint as TA4903, which has historically been known for its calculated and effective business email compromise (BEC) and phishing campaigns targeting various sectors, provides a case study in the diminishing returns of AI integration. Historically, TA4903 was successful because of its ability to impersonate government entities using highly tailored, believable lures. However, as the group has shifted toward using AI to generate these messages, the quality of their campaigns has deteriorated sharply.
The AI-generated lures often suffer from glaring structural issues, such as a complete lack of alignment between the subject lines, the content of the email, and the attached files. In some instances, the AI-generated emails are delivered with blank bodies, rendering the entire operation useless. Furthermore, the reliance on AI for coding has led to the deployment of "vibe-coded" phishing websites—sites that look and feel superficially similar to legitimate ones but are visually repetitive, simplistic, and easily flagged by both security filters and observant users.

Perhaps most counterintuitively, the obsession with "polishing" communications via AI has actually made some phishing attempts less believable. Humans, in their day-to-day communications, are inherently imperfect. Real-world emails from colleagues or partners often contain minor grammatical errors, typos, and a lack of formal structure. By using AI to create overly perfect, syntactically flawless messages, attackers are inadvertently creating a "tell" that security-aware employees can identify.
"Human beings often don’t know how to spell, they often are bad at grammar, and people aren’t trying to be overly polished," Larson noted. "I think that’s actually a bit of a detriment to some of these actors."
Exploits and the Limits of Automation
The limitations of AI are not confined to simple phishing lures; they extend into the more technical domain of vulnerability exploitation. While it is well-documented that AI can assist in the rapid discovery of software flaws, the actual exploitation process remains a high-stakes, precision-based endeavor.
Larson pointed to the "BlueMoon" exploit chain, which emerged in August 2026. This campaign involved the chaining of three separate vulnerabilities within the Microsoft Chrome browser. The threat actors behind the campaign utilized LLMs to construct their exploit code, relying on public patch release notes as their primary source of information.
Despite the technical sophistication required to chain these vulnerabilities, the AI-generated exploit was deeply flawed. It failed to behave with the reliability expected of professional-grade exploit kits and produced a significant amount of noise during execution. For defenders, this noise acts as a beacon, making the attack far easier to detect, trace, and mitigate than a more refined, human-crafted exploit would have been.
Defensive Priorities Remain Unchanged
The implications of these findings are profound for the cybersecurity community. If AI is not the magical lever that elevates every threat actor to a tier-one status, then the panic surrounding its adoption may be largely misplaced. Larson emphasized that AI is, fundamentally, just another productivity tool—one that is strictly limited by the skill of the operator.
"AI is only as good as its operator, and it can lead to more sloppy mistakes," Larson reiterated.
This perspective serves as a call for a return to fundamentals. While the industry has been racing to integrate AI into defensive stacks, the core requirements of cybersecurity—visibility, patch management, incident response, and user awareness—remain the most effective deterrents against the vast majority of threats, regardless of whether those threats are AI-generated or not.
The shift that is actually occurring is not necessarily in the sophistication of the attacks, but rather in the velocity of the threat. AI is undeniably making it easier for low-skill actors to attempt to exploit well-known, long-standing vulnerabilities. However, because the tools themselves are prone to error and the generated artifacts are often noisy, the actual risk profile for organizations may not have shifted as dramatically as once feared.
As the cybersecurity industry looks toward the coming years, the takeaway from Proofpoint’s analysis is one of cautious pragmatism. "Attacks are attacks, and AI generally doesn’t really matter," Larson concluded. "Whatever you’re defending against, defense is going to remain the same." For organizations, this means that the best investment remains in robust security hygiene rather than a reactionary race to deploy AI-centric solutions to fight a threat that, in many cases, is effectively sabotaging itself.
