While the cybersecurity industry has spent years anticipating that artificial intelligence would provide threat actors with advanced capabilities, recent research suggests that overreliance on these tools is frequently degrading the quality of malicious campaigns. Rather than creating a new class of sophisticated threats, the integration of AI into attack chains has often resulted in technical errors, operational inconsistencies, and more easily detectable lures.
Selena Larson, a principal threat researcher at Proofpoint, presented findings during the IRISSCON 2026 conference in Dublin on November 11, challenging the narrative that AI automatically elevates the efficacy of cyberattacks. According to Larson, the data indicates that AI is being used as a productivity tool that, when used without sufficient human oversight, introduces a phenomenon described as AI slop. This manifests as broken code, mismatched email components, and overly polished content that fails to mimic human behavior.
Proofpoint’s analysis of recent malware campaigns has identified instances where AI-generated code is functionally broken. In these cases, the code often contains comments that contradict the actual operations the malware performs. This disconnect suggests that attackers are utilizing AI to generate payloads without performing the necessary verification to ensure the scripts function as intended.
The impact of this reliance is also visible in phishing operations. Proofpoint highlighted the activities of a threat actor tracked as TA4903, a group known for business email compromise and phishing. Historically, the group utilized sophisticated lures to impersonate government entities. However, since incorporating AI into their workflow, the quality of their campaigns has declined. Researchers observed that AI-generated emails often feature blank message bodies or content that does not align with the provided attachments, making the attempts more obvious to targets.
Furthermore, the use of AI to generate phishing websites has led to a standardized, simplistic aesthetic that makes these sites easier for security teams to identify. Additionally, the AI’s tendency to produce grammatically perfect, highly polished text can serve as a disadvantage for attackers. Because human communication is often characterized by minor errors and inconsistent phrasing, the overly formal tone generated by AI can act as a red flag, making the phishing attempts appear less authentic.
In the realm of vulnerability exploitation, AI has also shown limitations. Larson cited the BlueMoon exploit chain, first observed in August 2026, which targeted three Microsoft Chrome browser vulnerabilities. Although the threat actors utilized large language models to draft the exploits based on patch release notes, the resulting code was flawed. The AI-generated exploits failed to behave as intended and created significant digital noise, which allowed defenders to identify and mitigate the attempts more readily.
Despite these failures, researchers confirm that AI has successfully increased the speed and scalability of certain attack phases, particularly regarding the exploitation of known vulnerabilities. However, this does not equate to an increase in overall sophistication. The consensus from these findings is that AI is only as effective as its operator; when used to bypass manual effort, it frequently introduces preventable mistakes that simplify the task for security professionals.
Ultimately, the core requirements for cyber defense remain unchanged. While attackers are leveraging AI to automate parts of their process, the fundamental nature of the threats has not shifted significantly. Defenders are advised to maintain their existing security postures, as the reliance on AI by threat actors has not fundamentally altered the threat landscape or necessitated a change in defense priorities.




