Gaslight Malware Targets AI-Powered Analysis on macOS

Researchers have uncovered Gaslight malware, a new macOS threat that targets AI-powered malware analysis tools instead of traditional security defenses. The malware hides fake error messages and prompt injection content inside its code to confuse automated analysis systems and disrupt investigations.

The discovery shows that cybercriminals have started adapting their malware to exploit the growing use of artificial intelligence in cybersecurity.

Gaslight Malware Targets AI Analysis

Researchers linked Gaslight malware to a North Korean threat actor with high confidence. Unlike many advanced malware families, Gaslight does not rely on new evasion techniques or sophisticated exploits.

Instead, the attackers focus on the analysis process itself. They attempt to manipulate AI assistants that security researchers use during malware investigations.

The developers wrote the malware in Rust and equipped it with backdoor and information-stealing capabilities. Those features closely resemble previous malware attributed to the same threat group.

Fake System Messages Mislead AI Models

Gaslight stands out because it embeds a 3.5 KB payload containing 38 fake system messages.

The malware inserts fabricated developer logs, debugging output, crash reports, and application alerts directly into the executable. Many of the messages use Markdown formatting and template-style placeholders to make them appear authentic.

Researchers found fake memory dumps, token expiration notices, Redis connection failures, SQL injection alerts, JSON parsing errors, build failures, and other messages that have no connection to the malware’s actual behavior.

The attackers created these messages to influence AI models that inspect malware samples during automated analysis.

Prompt Injection Becomes an Anti-Analysis Tool

Researchers believe the malware uses prompt injection to manipulate large language models rather than evade security sandboxes.

The embedded messages suggest that the AI session has failed or produced unreliable results. Some warnings claim authentication tokens have expired. Others report out-of-memory crashes, disk failures, or repeated processing errors.

By presenting false diagnostic information, the attackers hope the AI assistant will stop the analysis, shorten its report, or refuse to continue.

Researchers named the malware Gaslight because it attempts to make AI systems question their own conclusions instead of hiding its malicious behavior.

A New Challenge for AI-Powered Security

Traditional malware often tries to bypass antivirus software or escape virtual machines. Gaslight takes a different approach by attacking AI-assisted investigation workflows.

The researchers did not confirm that the technique can successfully defeat existing AI malware analysis platforms. However, they believe the campaign demonstrates a growing interest in AI-specific anti-analysis methods.

As security teams continue integrating large language models into malware research, attackers will likely keep developing techniques that manipulate automated analysis instead of conventional detection systems.

The emergence of Gaslight malware highlights a new phase in the ongoing battle between cybercriminals and defenders, where artificial intelligence has become another target rather than just another security tool.