Cybersecurity researchers at **SentinelOne** have identified a novel macOS malware family, **macOS.Gaslight**, which employs a unique strategy to bypass AI-assisted malware analysis. Instead of merely evading sandbox execution, **Gaslight** aims to sow confusion directly within AI systems tasked with dissecting its code. ### The 'Gaslighting' Mechanism The **Gaslight** malware is a **Rust** binary equipped with typical backdoor and information-stealing functionalities. However, its standout feature is a 3.5 KB payload embedded within the executable, containing 38 fabricated "system" messages. These messages are meticulously crafted to resemble legitimate developer logs, crash reports, debugging output, and program alerts, often utilizing Markdown formatting and template-style placeholders. Examples of these deceptive strings include: These fake messages reference a variety of non-existent issues, such as memory dumps, token expiration warnings, **Redis** connection failures, build-pipeline errors, and **SQL** injection alerts – all unrelated to the malware's actual behavior. ### Targeting AI Perception, Not Sandboxes **SentinelOne** emphasizes that the primary goal of these embedded strings is not to evade execution within a sandbox environment. Instead, they are designed as prompt injection content to make a Large Language Model (LLM)-assisted triage agent question the validity of its own analytical session or even refuse to continue analysis. "Its most notable feature is an embedded cascade of fabricated system-failure messages, designed to make an LLM-assisted triage agent doubt its own session," explains **SentinelOne**. "It attacks the agent's perception, rather than the sandbox it runs in. Accordingly, we dub this family **macOS.Gaslight**." ### A New Frontier in Anti-Analysis While **SentinelOne** did not definitively demonstrate that this technique could successfully bypass existing AI malware analysis platforms, its discovery signals a significant shift in threat actor methodologies. It indicates that adversaries are actively experimenting with and developing anti-analysis techniques specifically tailored to counteract AI-assisted security solutions. Attribution for **macOS.Gaslight** points with high confidence to a North Korean-linked threat actor, underscoring the sophisticated nature of state-sponsored cyber operations and their continuous innovation in evading detection.