North Korean state-sponsored actors have deployed MacOS.Gaslight, a Rust-based information stealer and backdoor targeting macOS environments. The implant utilizes a novel evasion technique by embedding 38 adversarial prompt injection strings designed to manipulate LLM-based malware triage tools. By targeting the cognitive layer of analysis, the malware attempts to trigger AI safety guards or provide fabricated system context, inducing AI assistants to misclassify the payload as benign or refuse analysis. This strategy directly degrades the accuracy of AI-assisted SOC triage, increasing attacker dwell time by blinding automated security analysis pipelines.

• Campaign Overview & Attribution

  • Attributed with high confidence to DPRK state-sponsored threat actors.
  • Targets high-value macOS endpoints using a specialized Rust-based implant.
  • Represents a paradigm shift from traditional code obfuscation to "AI-aware" cognitive evasion.

• Attack Mechanics & Prompt Injection

  • Embeds 38 specific adversarial strings within the binary's data sections to target LLM triage agents.
  • Attempts to trigger AI "safety" guardrails to force a refusal of analysis.
  • Provides fabricated system context to trick LLMs into summarizing the malware as a benign system utility.

• Technical Implementation (Rust Implant)

  • Developed in Rust to leverage memory safety and complicate analysis for legacy reverse-engineering tools.
  • Functions as a multi-purpose information stealer and persistent backdoor.
  • Includes anti-analysis payloads specifically tailored to mislead AI-assisted sandboxes.

• Security & Systemic Impact

  • Introduces "cognitive blinding," where security analysts rely on corrupted or misleading AI summaries.
  • Increases the risk of false negatives in SOC environments heavily dependent on LLM-driven summarization.
  • Serves as a scalable proof-of-concept for manipulating human-AI interaction during incident response.

• Mitigation & Defensive Actions

  • Implement mandatory human verification of AI-generated triage reports against raw binary strings and behavior.
  • Update LLM system prompts in triage pipelines to detect and ignore embedded adversarial instructions.
  • Deploy behavioral detection and EDR signatures focusing on unauthorized Rust-based persistence on macOS.